Inside and outside cabin double-temperature temperature compensation salinity-depth meter

Abstract

The invention provides an in-cabin and out-cabin dual-temperature compensation temperature-salinity-depth instrument, which belongs to the technical field of ocean measurement devices and comprises a control chip, a temperature sensing device, a pressure-depth sensing device, a conductivity measurement device, a data storage device and a communication device. The system measures the temperature inside and outside the cabin simultaneously through the inner and outer temperature sensors, dynamically adjusts compensation parameters by combining an adaptive temperature difference compensation function, and realizes accurate temperature measurement. And through an OceanParamNet deep learning model, the anomaly detection and parameter prediction capabilities are enhanced. The model is based on a hybrid architecture of a variational auto-encoder and a bidirectional long-short-term memory network, and introduces a sparse attention mechanism and physical constraint regularization. The system can adaptively adjust the sampling frequency according to the depth change rate, optimizes the power consumption, guarantees the data quality, and effectively solves the technical problem that the marine parameter measurement precision is not stable under the dynamic change condition of the temperature difference inside and outside the cabin.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine surveying devices, and more particularly, relates to a double-temperature-compensated thermosalinograph for both inside and outside the cabin. Background Art

[0002] A thermosalinograph is an important instrument used in oceanographic research to measure key parameters such as seawater temperature, salinity, and depth, and is widely used in fields such as ocean scientific expeditions, environmental monitoring, and climate change research. Traditional thermosalinographs usually use a single temperature sensor to measure seawater temperature, and measure depth and conductivity through a pressure sensor and a conductivity sensor respectively, and calculate salinity by combining empirical formulas. Such instruments can obtain relatively accurate measurement results under stable environmental conditions, providing basic data for the vertical profile analysis of ocean parameters.

[0003] However, traditional thermosalinographs often face the problem of measurement errors caused by the temperature difference change between the inside and outside of the instrument cabin during actual ocean observations. The temperature inside the cabin is affected by the heat generation of electronic components and environmental changes, and there is a difference from the actual seawater temperature outside the cabin, and this difference changes dynamically with depth and measurement time. Existing instruments mostly use fixed compensation parameters or simple linear correction methods to deal with the influence of temperature difference, and it is difficult to adapt to the complex and changeable ocean environment. Especially during the thermocline and rapid profile measurement processes, the measurement accuracy drops significantly.

[0004] The core problem that the current technology is difficult to solve is how to achieve high-precision, stable and reliable measurement of thermosalinograph parameters in a complex ocean environment with dynamic temperature difference changes between the inside and outside of the cabin. The traditional fixed-parameter compensation method cannot cope with the temperature difference change characteristics under different depths and different sea area conditions, and the simple mathematical model lacks the ability to describe the complex correlation relationship between parameters, resulting in systematic deviation in the measurement results and affecting the accuracy and reliability of ocean scientific research. That is to say, there is a technical problem that the measurement accuracy of the marine thermosalinograph is unstable under the condition of temperature difference between the inside and outside of the cabin in the existing technology. Summary of the Invention

[0005] In view of this, the present invention provides a double-temperature-compensated thermosalinograph for both inside and outside the cabin, which can solve the technical problem that the measurement accuracy of the marine thermosalinograph is unstable under the condition of temperature difference between the inside and outside of the cabin in the existing technology.

[0006] The present invention is implemented as follows: The present invention provides a dual-temperature compensated thermosalinograph for inside and outside the cabin. The temperature sensing device includes an internal temperature sensor and an external temperature sensor, which are used to measure the temperature inside the cabin and the temperature outside the cabin simultaneously and calculate the temperature difference compensation value; the ocean parameter measurement and control module is used to perform self-check, read system calibration parameters, collect temperature data, collect pressure data, collect conductivity data, data quality assessment, dynamically adjust temperature difference compensation parameters, adaptively adjust the sampling frequency, calculate seawater parameters, and upload measurement data steps; the ocean parameter measurement and control module calls the data quality assessment function and the OceanParamNet model in the data quality assessment step to comprehensively analyze the temperature inside the cabin, the temperature outside the cabin, pressure, depth, conductivity, and salinity data.

[0007] Among them, the measurement accuracy of the internal temperature sensor is ±0.01 °C, the measurement accuracy of the external temperature sensor is ±0.005 °C, and the sampling frequency of the temperature sensing device is 2 times per second.

[0008] Among them, the measurement range of the pressure-depth sensing device is 0 - 1000 meters, the measurement accuracy is ±0.1% full scale, and the sampling frequency of the pressure-depth sensing device is 1 time per second.

[0009] Among them, the conductivity measurement device uses the four-electrode method for measurement, the measurement range is 0 - 70 mS / cm, the measurement accuracy is ±0.005 mS / cm, and the sampling frequency of the conductivity measurement device is 1 time per second.

[0010] Among them, the inputs of the adaptive temperature difference compensation function include the temperature value inside the cabin measured by the temperature sensing device, the original temperature value outside the cabin measured by the temperature sensing device, the historical temperature difference data sequence stored in the data storage device, the depth value of the instrument measured by the pressure-depth sensing device, and the temperature change rate calculated by the temperature sensing device; the output of the adaptive temperature difference compensation function is the corrected temperature value outside the cabin after precise compensation and the updated value of the compensation coefficient.

[0011] Among them, the adaptive temperature difference compensation function adopts the piecewise polynomial fitting method, applies different compensation models for different temperature ranges and depth conditions, and dynamically adjusts the compensation parameters according to the statistical characteristics of the historical temperature difference data.

[0012] Among them, the adaptive temperature difference compensation function internally implements an adaptive weight distribution mechanism, uses a lower weight coefficient for the temperature mutation region to reduce the influence of abnormal data on the compensation effect, and at the same time realizes the self-adjustment of the compensation model through cumulative error analysis to ensure the stability and reliability of the compensation accuracy during the long-term measurement process.

[0013] Among them, the data quality evaluation function identifies abnormal data points through multi-dimensional parameter correlation analysis and time-series consistency test. The OceanParamNet model further enhances the anomaly detection ability through its variational autoencoder branch to mark or eliminate abnormal data. The input of the data quality evaluation function includes the in-cabin temperature data measured by the temperature sensing device, the out-of-cabin temperature data measured by the temperature sensing device, the pressure data measured by the pressure-depth sensing device, the depth data calculated by the pressure-depth sensing device, the conductivity data measured by the conductivity measurement device, and the salinity data calculated according to the practical salinity standard formula. The output of the data quality evaluation function is the data quality marking result.

[0014] Among them, the OceanParamNet model is an ocean parameter correlation deep learning model. The specific structure is a hybrid architecture based on variational autoencoder and bidirectional long short-term memory network. The front end uses a multi-layer perceptron to extract features from the original measurement data. The middle layer is composed of a bidirectional long short-term memory network. The back end is divided into two branches. One branch reconstructs the input data through a variational autoencoder for anomaly detection, and the other branch predicts the derivative parameters of seawater density and sound speed through a deep fully connected network. The OceanParamNet model introduces a sparse attention mechanism, and the attention window size of the sparse attention mechanism is dynamically adjusted according to the sampling frequency of the temperature sensing device and the seawater depth change rate measured by the pressure-depth sensing device to ensure that more computing resources are allocated in the key change areas.

[0015] Among them, the data quality evaluation function also includes a gradient analysis module, which identifies physically impossible mutations by checking the parameter change rate and designs different marking mechanisms for different types of anomalies to provide an accurate quality control basis for subsequent data processing.

[0016] Compared with the prior art, the present invention provides an in-cabin and out-of-cabin dual-temperature compensated thermosalinograph. The present invention proposes an in-cabin and out-of-cabin dual-temperature compensated thermosalinograph, which realizes high-precision ocean parameter measurement by simultaneously measuring the in-cabin and out-of-cabin temperatures and dynamically adjusting the temperature difference compensation coefficient in combination with an adaptive temperature difference compensation function. The design of internal and external dual-temperature sensors is adopted. The accuracy of the internal temperature sensor reaches ±0.01 °C, and the accuracy of the external temperature sensor reaches ±0.005 °C. It is combined with a high-precision pressure-depth sensing device and a conductivity measurement device to build a complete ocean parameter measurement platform.

[0017] Compared with traditional technologies, the present invention breaks through the limitations of fixed-parameter compensation and introduces an adaptive temperature difference compensation function, a data quality evaluation function, and an OceanParamNet deep learning model to form a complete solution. The adaptive temperature difference compensation function can dynamically adjust compensation parameters according to historical temperature difference data, the current temperature change rate, and environmental conditions; the data quality evaluation function identifies abnormal data through multi-dimensional parameter correlation analysis and time-series consistency tests; the OceanParamNet model uses variational autoencoders and bidirectional long short-term memory networks to capture complex non-linear relationships between parameters, further enhancing the ability of anomaly detection and parameter prediction.

[0018] The present invention successfully solves the technical problem of unstable measurement accuracy under dynamic changes in the temperature difference between the inside and outside of the cabin. It not only significantly improves the accuracy of seawater temperature, salinity, and depth measurements but also realizes the adaptive optimization of the measurement process through intelligent algorithms, providing more reliable technical support for oceanographic research and promoting the progress of ocean observation technology. Brief Description of the Drawings

[0019] Figure 1 It is a schematic diagram of the composition of the CTD instrument of the present invention.

[0020] Figure 2 It is a flowchart of the method of the ocean parameter measurement control module in the present invention. Detailed Embodiments

[0021] To make the purpose, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0022] As Figure 1As shown in the figure, it is a schematic diagram of the composition of a dual-temperature compensated thermosalinograph inside and outside the cabin provided by the present invention, including a control chip, a temperature sensing device, a pressure-depth sensing device, a conductivity measurement device, a data storage device, and a communication device. The control chip is electrically connected to the temperature sensing device, the pressure-depth sensing device, the conductivity measurement device, the data storage device, and the communication device respectively. A marine parameter measurement and control module is provided in the control chip. The temperature sensing device is used to measure the temperature inside and outside the cabin simultaneously and calculate the temperature difference compensation value. The pressure-depth sensing device is used to measure the seawater pressure and convert it into a depth value. The conductivity measurement device is used to measure the seawater conductivity and calculate the salinity value in combination with the temperature. The data storage device is used to record all measurement data and system calibration parameters. The communication device is used to realize communication with the upper computer and data transmission functions. The temperature sensing device includes an internal temperature sensor and an external temperature sensor. The measurement accuracy of the internal temperature sensor is ±0.01°C, and the measurement accuracy of the external temperature sensor is ±0.005°C. The sampling frequency of the temperature sensing device is 2 times per second. The measurement range of the pressure-depth sensing device is 0-1000 meters, and the measurement accuracy is ±0.1% full scale. The sampling frequency of the pressure-depth sensing device is 1 time per second. The conductivity measurement device uses the four-electrode method for measurement, with a measurement range of 0-70 mS / cm and a measurement accuracy of ±0.005 mS / cm. The sampling frequency of the conductivity measurement device is 1 time per second. For the convenience of description, the dual-temperature compensated thermosalinograph inside and outside the cabin of the present invention is hereinafter referred to as the system.

[0023] As Figure 2 shown, the marine parameter measurement and control module is used to perform the following steps:

[0024] S01. After the system is powered on, it first performs self-check. The working states of the internal temperature sensor and the external temperature sensor are detected through the temperature sensing device. The working state of the pressure sensor is detected through the pressure-depth sensing device. The electrode state of the conductivity sensor is detected through the conductivity measurement device. The detection results are stored in the data storage device.

[0025] S02. Read the system calibration parameters from the data storage device, including temperature calibration coefficient, pressure calibration coefficient, conductivity calibration coefficient, and depth conversion coefficient, and apply them to subsequent measurement calculations.

[0026] S03. Simultaneously collect the temperature inside and outside the cabin through the temperature sensing device, call the adaptive temperature difference compensation function to calculate the temperature difference value and obtain the corrected value of the temperature outside the cabin. The adaptive temperature difference compensation function dynamically adjusts the temperature difference compensation coefficient based on historical temperature difference data, current temperature change rate, and environmental conditions. The original temperature data and the corrected temperature data are stored in the data storage device.

[0027] S04. Collect seawater pressure data through the pressure-depth sensing device, correct it using the pressure calibration coefficient read in step S02, convert the pressure to a depth value according to the corrected pressure value combined with the current latitude position information through the international seawater state equation, and store the original pressure data and the calculated depth data in the data storage device;

[0028] S05. Collect seawater conductivity data through the conductivity measurement device, correct it using the conductivity calibration coefficient read in step S02, calculate the salinity value through the practical salinity standard formula in combination with the corrected outside-cabin temperature obtained in step S03 and the depth data calculated in step S04, and store the original conductivity data and the calculated salinity data in the data storage device;

[0029] S06. Call the data quality assessment function and the OceanParamNet model to comprehensively analyze the in-cabin temperature, outside-cabin temperature, pressure, depth, conductivity, and salinity data obtained in steps S03 to S05. The data quality assessment function identifies abnormal data points through multi-dimensional parameter correlation analysis and time-series consistency test. The OceanParamNet model further enhances the abnormal detection ability through its variational autoencoder branch, marks or eliminates the abnormal data, and improves the quality of the measurement data;

[0030] S07. Based on the updated value of the compensation coefficient output by the adaptive temperature difference compensation function in step S03 and the abnormal data correction suggestions provided by the data quality assessment function in step S06, evaluate the impact of the temperature change inside and outside the cabin on the measurement results in real time, and dynamically adjust the temperature difference compensation parameters to ensure stable measurement accuracy under different environmental conditions;

[0031] S08. According to the change rate of the seawater depth measured in step S04, adaptively adjust the sampling frequencies of the pressure-depth sensing device and the conductivity measurement device, increase the sampling frequency in the rapidly changing area and decrease the sampling frequency in the stable area to optimize the system power consumption, and apply the adjusted sampling frequencies to the data acquisition processes in steps S03 to S05;

[0032] S09. Call the depth fully connected network branch of the OceanParamNet model, calculate the seawater density, sound speed, and other ocean physical parameters based on the seawater temperature-salinity-depth data obtained in steps S03 to S05, construct a complete ocean parameter profile, and store the calculation results in the data storage device to provide data support for oceanographic research;

[0033] S10. Regularly upload the measurement data obtained and calculated in steps S01 to S09 to the host computer through the communication device. At the same time, receive the parameter settings and control instructions sent by the host computer, adjust the system working mode and measurement parameters according to the received instructions, and apply the new parameters to the execution process of steps S03 to S09.

[0034] The adaptive temperature difference compensation function is used to optimize the temperature difference compensation algorithm in step S03 and improve the measurement accuracy of the external cabin temperature. The adaptive temperature difference compensation function dynamically adjusts the temperature difference compensation coefficient by analyzing historical temperature difference data, the current temperature change rate, and environmental conditions to achieve more accurate correction of the external cabin temperature. The inputs of the adaptive temperature difference compensation function include the cabin temperature value measured by the temperature sensing device, the original external cabin temperature value measured by the temperature sensing device, the historical temperature difference data sequence stored in the data storage device, the depth value of the instrument measured by the pressure-depth sensing device, and the temperature change rate calculated by the temperature sensing device; the outputs of the adaptive temperature difference compensation function are the corrected external cabin temperature value after precise compensation and the compensation coefficient update value for step S07. The adaptive temperature difference compensation function adopts a piecewise polynomial fitting method, applies different compensation models for different temperature ranges and depth conditions, and dynamically adjusts the compensation parameters according to the statistical characteristics of the historical temperature difference data. The adaptive temperature difference compensation function internally implements an adaptive weight allocation mechanism, uses a lower weight coefficient for the temperature mutation region to reduce the influence of abnormal data on the compensation effect, and at the same time realizes self-adjustment of the compensation model through cumulative error analysis to ensure the stability and reliability of the compensation accuracy during the long-term measurement process.

[0035] The data quality assessment function is used to optimize the abnormal data detection process in step S06 and improve the system's data processing ability. The data quality assessment function identifies and processes abnormal data points that occur during the measurement process through multi-dimensional parameter correlation analysis and temporal consistency verification. The inputs of the data quality assessment function include the in-cabin temperature data measured by the temperature sensing device, the out-of-cabin temperature data measured by the temperature sensing device, the pressure data measured by the pressure-depth sensing device, the depth data calculated by the pressure-depth sensing device, the conductivity data measured by the conductivity measurement device, and the salinity data calculated according to the practical salinity standard formula; the output of the data quality assessment function is the data quality marking result and the abnormal data correction suggestion. The data quality marking result is stored in the data storage device, and the abnormal data correction suggestion is used for parameter adjustment in step S07. The data quality assessment function first calculates the correlation matrix between parameters, then determines the normal correlation range according to the historical statistical model, then uses the Mahalanobis distance algorithm to detect abnormal points in the multi-dimensional parameter space, and finally applies the sliding window analysis technology to evaluate the temporal coherence and comprehensively judge the reliability level of the data points. The data quality assessment function also includes a gradient analysis module, which identifies physically impossible mutations by checking the parameter change rate and designs different marking mechanisms for different types of abnormalities, providing an accurate quality control basis for subsequent data processing.

[0036] The OceanParamNet model is used to optimize steps S06 and S09, improve the abnormal data detection ability and enhance the accuracy of ocean parameter calculation. The OceanParamNet model is a deep learning model for ocean parameter association, which can learn complex non-linear relationships from multi-dimensional ocean parameter data to achieve abnormal detection and parameter prediction.

[0037] The specific structure of the OceanParamNet model is a hybrid architecture based on a variational autoencoder and a bidirectional long short-term memory network. At the front end of the OceanParamNet model, a multi-layer perceptron is used to extract features from the original measurement data such as the temperature measured by the temperature sensing device, the pressure measured by the pressure-depth sensing device, and the conductivity measured by the conductivity measurement device. The middle layer of the OceanParamNet model is composed of a bidirectional long short-term memory network, which is used to capture the characteristic of parameter time series change. The back end of the OceanParamNet model is divided into two branches. One branch reconstructs the input data through a variational autoencoder for anomaly detection in step S06. The other branch predicts the derived parameters such as seawater density and sound speed in step S09 through a deep fully connected network. The OceanParamNet model introduces a sparse attention mechanism, and the attention window size of the sparse attention mechanism is dynamically adjusted according to the sampling frequency of the temperature sensing device and the seawater depth change rate measured by the pressure-depth sensing device, ensuring that more computing resources are allocated in the key change areas. The OceanParamNet model also includes an uncertainty estimation module, which provides a confidence interval for each prediction result to assist the ocean parameter measurement control module to identify the prediction results that may have large errors. The OceanParamNet model adopts a residual connection structure to alleviate the difficulty of deep network training, and applies batch normalization technology to improve the training stability. The last layer of the OceanParamNet model adopts physical constraint regularization to ensure that the prediction results conform to the basic physical laws of oceanography.

[0038] The steps for establishing the training data set during the pre-training process of the OceanParamNet model specifically include: First, extract high-quality temperature-salinity-depth profile data from the global ocean observation database, covering different sea areas, seasons and depth ranges; then perform quality control on the original data, including outlier detection, data interpolation and standardization processing; then synthesize and expand the data set according to the physical ocean model to increase the sample proportion of extreme conditions and rare events; subsequently, simulate sensor noise and drift through the Monte Carlo method to generate measurement data containing various error situations; finally, divide the processed data into an 80% training set, a 10% validation set and a 10% test set, and ensure that each subset has similar statistical distribution characteristics. The training data set also includes the measurement results under different instrument models and different calibration parameters to enhance the generalization ability and adaptability of the OceanParamNet model. To simulate the actual application scenario, scenario data simulating the temperature difference inside and outside the simulation cabin is added to the training data set, and the influence degree of the temperature difference on the measurement results is marked, so that the OceanParamNet model can learn the internal law of temperature difference compensation.

[0039] The steps for pre-training the OceanParamNet model specifically include: in the first stage, the variational autoencoder part is pre-trained in an unsupervised learning manner, enabling the OceanParamNet model to learn the intrinsic distribution characteristics of ocean parameters; in the second stage, the bidirectional long short-term memory network is trained using labeled data to optimize the ability to extract temporal features; in the third stage, end-to-end joint training is carried out to integrate the variational autoencoder and the bidirectional long short-term memory network, and a multi-task learning objective is introduced to optimize both anomaly detection and parameter prediction performance simultaneously; in the fourth stage, through contrastive learning techniques, the sensitivity of the OceanParamNet model to parameter changes under similar ocean environmental conditions is enhanced; in the final stage, transfer learning methods are applied to fine-tune the OceanParamNet model using data from a small-scale target sea area to improve its performance in the target application area. During the training process, a learning rate decay strategy and an early stopping mechanism are adopted to prevent overfitting, and the AdamW optimizer is used to accelerate convergence. The performance evaluation of the OceanParamNet model adopts a multi-index comprehensive evaluation method, including root mean square error, anomaly detection F1 score, and physical consistency score. After the pre-training is completed, the parameters of the OceanParamNet model are saved for subsequent deployment and application in the in-cabin and out-of-cabin dual-temperature conductivity-temperature-depth profiler, and an incremental learning interface is retained to support the system in continuously accumulating experience and optimizing performance during actual application.

[0040] Among them, the variational autoencoder is a generative neural network model used to learn the latent representation and probability distribution of data, achieving data compression and reconstruction; the bidirectional long short-term memory network is a special recurrent neural network structure that can consider both the forward and backward temporal information of data simultaneously to improve the ability to process sequence data; the sparse attention mechanism is a neural network computing technology that selectively focuses on the key parts of the input data, improving the model efficiency and performance by reducing the computational weights of non-critical regions; the physical constraint regularization is a technology that introduces physical laws as constraint conditions into the neural network training process to ensure that the model prediction results conform to basic physical laws.

[0041] Specifically, the conductivity-temperature-depth profiler provided by the present invention is composed of a control chip, a temperature sensing device, a pressure-depth sensing device, a conductivity measuring device, a data storage device, and a communication device. The control chip is electrically connected to the temperature sensing device, the pressure-depth sensing device, the conductivity measuring device, the data storage device, and the communication device respectively. The details are described as follows.

[0042] Optionally, the control chip is the core processing unit of the system. It adopts a high-performance 32-bit microprocessor with an in-built ocean parameter measurement and control module, which is responsible for the coordinated control and data processing of each functional module of the system. The control chip features a low-power design, with a working voltage of 3.3V. The clock frequency can be dynamically adjusted within the range of 10MHz to 200MHz according to the computing requirements. It has 256KB SRAM and 2MB flash memory built-in, supports a floating-point arithmetic unit, and can meet the real-time computing requirements of complex algorithms.

[0043] Optionally, the temperature sensing device includes an internal temperature sensor and an external temperature sensor, which are used to measure the temperature inside and outside the cabin respectively. The internal temperature sensor is installed inside the instrument electronic cabin, with a measurement accuracy of ±0.01°C and a temperature measurement range of -5°C to 45°C; the external temperature sensor is installed on the surface of the instrument housing, in direct contact with seawater, with a measurement accuracy of ±0.005°C and a temperature measurement range of -5°C to 35°C. Both temperature sensors use platinum resistance elements and are combined with a 24-bit high-precision analog-to-digital converter to achieve high-resolution temperature data acquisition, with a sampling frequency of 2 times per second.

[0044] Optionally, the pressure and depth sensing device uses a silicon resonant pressure sensor, with a measurement range of 0 - 1000 meters of water depth, equivalent to a pressure range of 0 - 10MPa, and a measurement accuracy of ±0.1% of full scale, that is, an error of ±1 meter in depth. This device integrates a temperature compensation circuit to effectively eliminate the influence of temperature changes on pressure measurement, and realizes high-precision digital signal output through a 16-bit analog-to-digital converter, with a sampling frequency of 1 time per second.

[0045] Optionally, the conductivity measurement device measures the conductivity of seawater using the four-electrode method. The electrode material is platinum, which has good seawater corrosion resistance. The measurement range is 0 - 70mS / cm, covering the conductivity change range of the main sea areas around the world, and the measurement accuracy is ±0.005mS / cm. This device is equipped with a signal conditioning circuit and a 24-bit analog-to-digital converter to achieve high-precision conductivity measurement, with a sampling frequency of 1 time per second.

[0046] Optionally, the data storage device uses a 32GB industrial-grade flash memory, organized in a circular storage structure, and can continuously record measurement data for about 6 months. The storage interface uses the SPI bus, supports high-speed data reading and writing, and realizes a data verification function to ensure data integrity. The data is stored in a timestamp index manner, which is convenient for subsequent data retrieval and analysis.

[0047] Optionally, the communication device includes two communication methods: an RS-485 interface and a Bluetooth Low Energy module. The RS-485 interface is used for wired connection to the host computer, supports long-distance stable communication, with a maximum transmission distance of up to 1200 meters, and the communication rate is adjustable from 9600 bps to 115200 bps; the Bluetooth Low Energy module is used for short-distance wireless communication, supports the BLE5.0 protocol, and the communication distance can reach 30 meters, facilitating rapid acquisition of on-site data and configuration of instrument parameters.

[0048] Optionally, each of the above hardware modules adopts a waterproof design, and the overall system is encapsulated in a titanium alloy shell, with a pressure-resistant depth of 1200 meters and an operating temperature range of -5°C to 40°C. The system is powered by a 12V lithium battery pack with a capacity of 10000 mAh, supporting continuous operation for more than 30 days. All circuit boards adopt a multi-layer design, separating the signal layer from the power layer, and are provided with good grounding and shielding measures to effectively suppress electromagnetic interference and improve system stability.

[0049] Optionally, the overall size of the in-cabin and out-of-cabin dual-temperature compensated CTD is a cylindrical structure with a diameter of 80 mm × a length of 300 mm, and the weight is about 1.5 kg, which is convenient for carrying various ocean observation platforms. When installing the system, it is necessary to ensure that the external temperature sensor and the conductivity electrode are in direct contact with seawater, while the pressure sensor is connected to the external seawater through a pressure conduit, so as to achieve accurate and reliable measurement of ocean parameters. Through this hardware design, combined with the adaptive temperature difference compensation algorithm and data processing method of the present invention, the technical problem of unstable measurement accuracy under the condition of dynamic change of the temperature difference between the inside and outside of the cabin is effectively solved.

[0050] The following provides a specific implementation manner of the steps executed by the ocean parameter measurement and control module.

[0051] The specific implementation of step S01 is to ensure the normal function of each sensing device through a systematic self-check process. After the system is powered on, the control chip first sends a self-check instruction to the temperature sensing device, and the temperature sensing device executes a test sequence internally to verify the working status of the internal temperature sensor and the external temperature sensor. The specific implementation is to send a standard test signal to each sensor, receive the feedback data, and compare the data with the preset normal range value. For the internal temperature sensor, the normal response time should be less than 50 milliseconds, and the signal fluctuation should not exceed ±0.005 °C; for the external temperature sensor, the normal response time should be less than 80 milliseconds, and the signal fluctuation should not exceed ±0.002 °C. Subsequently, the control chip sends a self-check command to the pressure and depth sensing device, and the pressure and depth sensing device detects the zero offset value and sensitivity coefficient of the pressure sensor through the internal circuit self-diagnosis program. The normal zero offset should be within ±0.05% of the full scale. The conductivity measurement device uses the impedance analysis method to detect the electrical connection status between the four electrodes. The impedance value between the electrodes should be within the range of 3 - 5 kΩ to be considered normal. All detection results are stored in the data storage device in a structured data format, including the detection timestamp, sensor identification code, status code, and detailed parameter values. This step provides a reliability guarantee for subsequent parameter measurements by establishing a reference state at system startup.

[0052] The specific implementation of step S02 is to accurately read and apply the calibration parameters. The control chip reads the calibration parameter table from the non-volatile storage area of the data storage device through the serial communication protocol. This table stores various calibration coefficients in a key-value pair structure, including the temperature calibration coefficient matrix (including slope and intercept values for linearly correcting temperature readings), the pressure calibration coefficient polynomial (usually third-order polynomial coefficients for non-linearly correcting pressure readings), the conductivity calibration coefficient matrix (including temperature-conductivity cross-correction terms), and the depth conversion coefficient (related to the local acceleration due to gravity). After reading, the control chip loads these parameters into the runtime parameter area of the ocean parameter measurement control module and performs parameter validity verification to ensure that all parameters are within a reasonable value range. The valid range of the temperature calibration coefficient is 0.9 - 1.1 times the nominal value, the valid range of the pressure calibration coefficient is 0.85 - 1.15 times the nominal value, and the valid range of the conductivity calibration coefficient is 0.9 - 1.1 times the nominal value. This step ensures high precision and traceability for subsequent measurement calculations by applying the verified calibration coefficients.

[0053] The specific implementation of step S03 is to achieve synchronous acquisition and correction of the internal and external cabin temperatures. The temperature sensing device simultaneously acquires the original values of the internal and external cabin temperatures according to a sampling frequency of 2 Hz. During the acquisition process, oversampling technology is adopted, that is, the actual sampling rate is 20 Hz, and it is reduced to 2 Hz for output through mean filtering to improve the signal-to-noise ratio. Subsequently, the adaptive temperature difference compensation function is called to correct the external cabin temperature. This function uses a multi-stage temperature difference compensation algorithm. First, it calculates the statistical characteristics based on the historical temperature difference data sequence (storing the most recent 3600 data points), including the average temperature difference value, standard deviation, and change trend. Then, according to the current temperature change rate (calculated by the temperature change amount within a 10-second window), the compensation coefficient is dynamically adjusted. When the temperature change rate is greater than 0.1 °C / minute, the fast response mode is adopted, and the compensation coefficient increases by 15%; when the temperature change rate is less than 0.02 °C / minute, the steady state mode is adopted, and the compensation coefficient returns to the standard value. In addition, environmental conditions such as depth changes will also affect the compensation parameters. For every 100-meter increase in water depth, the compensation coefficient is adjusted by approximately 2%. The final corrected value is calculated through a piecewise polynomial model, and different parameter groups are applied for different temperature ranges (such as -2 - 5 °C, 5 - 15 °C, 15 - 30 °C). The original temperature data and the corrected temperature data are stored in the data storage device in the form of a time series. This step improves the absolute accuracy of water temperature measurement by accurately compensating for the influence of the temperature difference between the inside and outside of the cabin.

[0054] The specific implementation of step S04 is to achieve precise processing of pressure measurement and depth conversion. The pressure-depth sensing device acquires seawater pressure data at a frequency of 1 Hz and corrects it using the pressure calibration coefficient read in step S02. The correction process adopts a third-order polynomial compensation model, which can effectively correct the non-linear error of the sensor. The corrected pressure value, combined with the current latitude position information (set through external input or default value), is used to perform depth conversion using the international seawater state equation. This equation takes into account the variation of seawater density with latitude, pressure, and temperature and is solved using the iterative approximation method. A local acceleration of gravity correction term is used in the calculation, and this term has a sine function relationship with latitude. For every 10-degree change in latitude, the correction value changes by approximately 0.5%. During the conversion process, a pressure hysteresis correction algorithm is also applied. This algorithm compensates for the hysteresis effect of the sensor by detecting the pressure difference during the rising and falling processes, and the correction value is usually 0.05 - 0.2% of the full scale. The original pressure data and the calculated depth data are stored in the data storage device in a timestamp-associated manner, and the parameter values used for the conversion are recorded simultaneously. This step ensures the accuracy and consistency of depth measurement under different sea area conditions.

[0055] The specific implementation of step S05 is to achieve high-precision processing of conductivity measurement and salinity calculation. The conductivity measurement device uses the four-electrode method to collect seawater conductivity data at a frequency of 1 Hz. The four-electrode method effectively eliminates the influence of electrode polarization effect and contact impedance by separating the excitation electrode and the detection electrode. The obtained original conductivity data is corrected using the conductivity calibration coefficient read in step S02. The correction process takes into account the non-linear variation of conductivity with temperature and uses a temperature-conductivity cross-calibration matrix that covers a temperature range of -2°C to 35°C and a conductivity range of 0 - 70 mS / cm, with an interpolation accuracy better than 0.002 mS / cm. The corrected conductivity, combined with the corrected out-of-cabin temperature obtained in step S03 and the depth data calculated in step S04, calculates the salinity value through the practical salinity standard formula. This formula is based on the seawater conductivity ratio method, takes into account the influence of pressure and temperature on conductivity, provides a reference value under standard seawater conditions (temperature of 15°C, pressure of 0 dbar), and then applies correction terms to adjust the salinity value under actual measurement conditions. The piecewise processing technique is used in the calculation process, dividing the complete calculation range into multiple sub-intervals, and each interval uses optimized coefficients to improve the calculation efficiency and accuracy. The original conductivity data and the calculated salinity data are stored in the data storage device in an associated record form. This step ensures the accuracy and reliability of salinity measurement in a complex marine environment.

[0056] The specific implementation of step S06 is to achieve a comprehensive assessment of multi-parameter data quality. This step calls the data quality assessment function and the OceanParamNet model to analyze the in-cabin temperature, out-cabin temperature, pressure, depth, conductivity, and salinity data obtained in steps S03 to S05. The data quality assessment function first calculates the Pearson correlation coefficient matrix between parameters and constructs an association model between parameters. Subsequently, based on historical statistical data, the normal correlation range is determined. For example, the correlation coefficient between temperature and salinity should usually be within the range of -0.7 to 0.3 in a specific sea area. Exceeding this range may indicate an anomaly. The function uses the Mahalanobis distance algorithm to detect outliers in the multi-dimensional parameter space, calculates the normalized distance of each data point to the distribution center, and marks it as an anomaly when the distance exceeds 3.5 standard deviations. At the same time, the sliding window technique (the window size is usually 11 - 21 data points) is applied to analyze the temporal coherence of parameters, calculate the local variance and mutation index, and mark it as a temporal anomaly when the mutation index exceeds the preset threshold (usually 2.8). The OceanParamNet model reconstructs the input data through its variational autoencoder branch, calculates the reconstruction error, and further confirms the anomaly status when the error exceeds the dynamic threshold (determined based on the historical error distribution, usually 3 times the average error). The abnormal data is divided into two categories: correctable and uncorrectable according to the severity. The correctable data is replaced by interpolation or model prediction, and the uncorrectable data is marked and excluded. All data quality assessment results are stored in the data storage device in a marked form. This step improves the reliability and scientific value of the measurement data through a method combining multi-dimensional analysis and machine learning.

[0057] The specific implementation of step S07 is to achieve the dynamic optimization and adjustment of the temperature difference compensation parameters. This step is based on the updated value of the compensation coefficient output by the adaptive temperature difference compensation function in step S03 and the abnormal data correction suggestions provided by the data quality evaluation function in step S06, and evaluates the impact of the temperature changes inside and outside the cabin on the measurement results in real time. The specific implementation is to construct a temperature difference impact evaluation model, which uses the recursive least squares method to analyze the correlation between the temperature difference changes and the measurement errors in the historical data. When the detected temperature change rate inside the cabin exceeds 0.2 °C / minute, the compensation parameter re-evaluation process is triggered. During the evaluation process, the absolute value of the temperature difference, the temperature difference change trend, and the frequency of abnormal data appearance are comprehensively considered to construct a weighted objective function. The optimization of this objective function uses the gradient descent method, with the step size adaptively adjusted, and the convergence condition is that the parameter change rate is less than 0.1% or the number of iterations reaches 50 times. The newly obtained compensation parameters are applied to the temperature difference compensation function in step S03 to form a closed-loop feedback mechanism. To prevent the parameters from fluctuating violently, a buffer mechanism is introduced, and the new parameters are smoothed using the exponentially weighted moving average method, with the smoothing coefficient usually being 0.15 - 0.3. At the same time, an upper limit for parameter changes is set, and the single adjustment amplitude does not exceed ±15% of the previous parameter. The optimized parameter values are recorded in the data storage device together with the timestamps for subsequent analysis. This step ensures the stability of the measurement accuracy in an environment with rapidly changing temperature conditions through the adaptive adjustment mechanism.

[0058] The specific implementation of step S08 is to achieve the adaptive adjustment control of the sampling frequency. This step dynamically adjusts the sampling frequencies of the pressure-depth sensing device and the conductivity measurement device according to the seawater depth change rate measured in step S04. The depth change rate is obtained by calculating the derivative of the depth value within a unit time, and the central difference method is used to improve the calculation accuracy. When the depth change rate is greater than 1.5 m / s, the system enters the fast response mode, and the sampling frequencies of the pressure-depth sensing device and the conductivity measurement device are increased to 2 times per second; when the change rate is between 0.5 - 1.5 m / s, the standard sampling frequency of 1 time per second is maintained; when the change rate is less than 0.5 m / s, the sampling frequency is reduced to 0.5 times per second. To prevent the sampling frequency from switching frequently, a hysteresis control mechanism is introduced, that is, the frequency adjustment is only triggered when the change rate continuously exceeds the threshold for more than 3 seconds. The change in the sampling frequency is notified to each sensing device through an interrupt, and the timer configuration in the control chip is updated. At the same time, the system power consumption management module adjusts the power distribution according to the current sampling frequency, and in the low-frequency sampling mode, some circuits enter the low-power state, with the power consumption reduced by up to 40%. The adjusted sampling frequency is applied to the data acquisition process in steps S03 to S05 in real time and is recorded in the system log. This step optimizes the system energy consumption and extends the device's battery life while ensuring the data density in the key areas through the intelligent adjustment of the sampling strategy.

[0059] The specific implementation of step S09 is to achieve the comprehensive calculation and construction of ocean parameter profiles. This step calls the depth fully connected network branch of the OceanParamNet model, and based on the seawater temperature, salinity, and depth data obtained in steps S03 to S05, calculates seawater density, sound speed, and other ocean physical parameters. The OceanParamNet model adopts a 5-layer fully connected network structure, with the number of nodes in each layer being 64, 128, 256, 128, and 64 respectively. The activation function uses LeakyReLU to prevent the problem of gradient disappearance. The model inputs include temperature, salinity, depth, and their first derivative values to capture the parameter change trends. In the calculation, a physical constraint layer is applied, which ensures that the prediction results satisfy the basic laws of thermodynamics through penalty terms, such as the increase of density with depth, the relationship between sound speed and temperature and salinity, etc. For the calculation of seawater density, the model output undergoes post-processing, combined with the correction terms of the international seawater equation of state to ensure accuracy under extreme conditions. The sound speed calculation uses a hybrid method, combining empirical formulas and neural network predictions. When the water depth is less than 200 meters, the weight of the empirical formula is 0.7, and as the depth increases, the neural network weight gradually increases to 0.9. The calculation results include the parameter values at standard depth layers (usually 0, 10, 20, 30, 50, 75, 100, 125, 150, 200, 250, 300, 400, 500, 600, 700, 800, 900, 1000 meters), and cubic spline interpolation is used. The finally constructed ocean parameter profile is stored in the data storage device in a structured format, including depth index, original measurement parameters, and calculated parameters. This step provides a comprehensive ocean parameter analysis through a method combining deep learning and physical models, providing high-quality data support for oceanographic research.

[0060] The specific implementation of step S10 is to achieve the closed-loop management of data transmission and system control. In this step, the measurement data obtained and calculated in steps S01 to S09 are regularly uploaded to the host computer through the communication device, and at the same time, the parameter settings and control instructions sent by the host computer are received. Optionally, a hierarchical transmission strategy is adopted for data upload. Key parameters (such as temperature, salinity, depth) are transmitted once every 30 seconds with high priority, derived parameters (such as density, sound speed) are transmitted once every 2 minutes with medium priority, and system status information is transmitted once every 5 minutes with low priority. The transmission encapsulates the data in JSON format, including timestamp, parameter identifier, value, unit, and quality flag. The communication process implements an automatic retransmission mechanism. When a transmission error is detected, the system automatically retransmits up to 3 times. When receiving instructions from the host computer, the communication device first verifies the instruction format and checksum, and then parses the instruction type, which is divided into two categories: parameter setting instructions and control instructions. Parameter setting instructions are used to adjust system working parameters, such as sampling frequency, calibration coefficient, etc.; control instructions are used to switch the system working mode, such as standard measurement mode, low power consumption mode, high-precision mode, etc. A legality check is performed before the instruction is executed to ensure that the parameters are within the safe range. The execution result is returned to the host computer through the status feedback mechanism, including the execution status code and detailed information. The new parameters and working modes are immediately applied to the execution process of steps S03 to S09 to achieve dynamic adjustment of system configuration. This step realizes the seamless integration of the system and the host computer by establishing a stable two-way communication mechanism, improving the flexibility and adaptability of the overall measurement system.

[0061] The following details the mathematical models or calculation processes involved in the present invention.

[0062] For the adaptive temperature difference compensation function in step S03, its core calculation process is specifically expressed as follows:

[0063] T ext_corrected =T ext_raw +ΔT comp ;

[0064] In the formula, T ext_corrected is the corrected outside-cabin temperature value, with the unit of °C; T ext_raw is the original temperature value measured by the outside-cabin temperature sensor, with the unit of °C; ΔT comp is the temperature difference compensation value, with the unit of °C.

[0065] The calculation of the temperature difference compensation value adopts a piecewise polynomial model:

[0066]

[0067] In the formula, a i is the coefficient of the temperature difference compensation polynomial, usually n is taken as 3; T int is the inside-cabin temperature, with the unit of °C; T refis the reference temperature, with a default value of 20°C; is an adjustment function related to the depth and the rate of temperature change; d is the current depth, in meters; is the rate of temperature change, in °C / minute.

[0068] Adjustment function Specifically, it is expressed as follows:

[0069]

[0070] In the formula, k d is the depth influence coefficient, with a default value of 0.02; k T is the rate of temperature change influence coefficient, with a default value of 0.15.

[0071] The polynomial coefficient a i is obtained by fitting historical data using the least squares method and has different values for different temperature ranges. The obtaining method is to collect a large number of calibration data points (T int , T ext_raw , T ext_true ), where T ext_true is the true external temperature measured by the reference thermometer, and then by solving the following system of equations:

[0072]

[0073] In the formula, m is the number of calibration data points; the superscript (j) represents the jth calibration data point.

[0074] Rate of temperature change is calculated by the central difference method:

[0075]

[0076] In the formula, Δt is the time step, in minutes, and the typical value is 1 / 6 minute (10 seconds).

[0077] The adaptive temperature difference compensation function is modeled by polynomial fitting, which can effectively describe the compensation relationship under different temperature difference conditions, and the exponential term can capture the non-linear characteristics. The adjustment functions of depth and rate of temperature change introduce the influence of environmental factors and improve the adaptability of the model. This system of equations combines an empirical model and a real-time adjustment mechanism, and the accuracy is improved by about 40% compared with a simple linear compensation model.

[0078] For the conversion process from pressure to depth in step S04, it is specifically expressed as follows:

[0079]

[0080] In the formula, P corrected is the corrected pressure value, in dbar; Praw is the original pressure value measured by the pressure sensor, with the unit of dbar; b i are the pressure calibration polynomial coefficients, which are obtained by calibration with a pressure standard device.

[0081] Depth conversion adopts the international seawater equation of state:

[0082]

[0083] where z is the depth, with the unit of meter; g(φ) is the local acceleration of gravity, which is related to the latitude φ; ρ(T, S, P corrected ) is the seawater density, with the unit of kg / m 3 ; T is the temperature, with the unit of °C; S is the salinity, with the unit of PSU; p is the pressure integration variable, with the unit of dbar.

[0084] Calculation formula for the local acceleration of gravity:

[0085] g(φ) = 9.780318·(1 + 5.2788×10 -3 sin 2 φ + 2.36×10 -5 sin 4 φ);

[0086] where φ is the latitude, with the unit of degree.

[0087] Due to the complexity of integral calculation, an approximate formula is usually adopted in practical applications:

[0088]

[0089] where γ is the depth conversion coefficient, and the typical value is 0.99, considering that the average seawater density is about 1028 kg / m 3 .

[0090] Pressure hysteresis correction algorithm:

[0091]

[0092] where P final is the final corrected pressure value, with the unit of dbar; δ h is the hysteresis coefficient, and the typical value range is 0.0005 - 0.002; is the pressure change direction function, which is 1 for rising and -1 for falling; is the pressure change rate, with the unit of dbar / second.

[0093] The pressure-to-depth conversion equation is based on the principle of hydrostatic equilibrium, taking into account the variations of seawater density with temperature, salinity, and pressure, as well as the inhomogeneity of the Earth's gravitational field. The approximate formula simplifies the calculation process and is applicable to real-time application scenarios. The hysteresis correction algorithm compensates for the mechanical hysteresis effect of the pressure sensor, improving the measurement consistency during the ascent and descent processes.

[0094] For the conductivity calibration and salinity calculation processes in step S05, they are specifically represented as follows:

[0095] Conductivity correction formula:

[0096] C corrected =C raw ·(1 + α·(T - T ref ) + β·(T - T ref ) 2 );

[0097] In the formula, C corrected is the corrected conductivity, with the unit of mS / cm; C raw is the original measured conductivity, with the unit of mS / cm; α is the first-order temperature coefficient, with a typical value of 0.02 / °C; β is the second-order temperature coefficient, with a typical value of 0.0005 / °C²; T is the current temperature, with the unit of °C; T ref is the reference temperature, with a typical value of 15°C.

[0098] Conductivity ratio calculation:

[0099]

[0100] In the formula, R is the conductivity ratio, dimensionless; C KCl is the conductivity of the standard KCl solution at the same temperature, with the unit of mS / cm.

[0101] Salinity calculation uses the practical salinity standard formula:

[0102] S = a 0 + a 1 R 1 / 2 + a 2 R + a 3 R 3 / 2 + a 4 R 2 + a 5 R 5 / 2 + ΔS(T, P);

[0103] In the formula, S is the practical salinity, dimensionless; a 0 to a 5is the standard coefficient, with values of 0.0080, -0.1692, 25.3851, 14.0941, -7.0261, and 2.7081 respectively; ΔS(T, P) is the temperature and pressure correction term.

[0104] Calculation of the temperature and pressure correction term:

[0105]

[0106] In the formula, b 0 to b 5 are the pressure correction coefficients; k is the temperature influence coefficient, with a typical value of 0.0162; n is the temperature normalization coefficient, with a typical value of 0.0116; P is the pressure, in dbar; T is the temperature, in °C.

[0107] The conductivity and salinity calculation equations are based on the relationship between seawater ion concentration and conductivity, and a polynomial fitting model is used to describe the nonlinear characteristics. The design of the exponential terms 0.5, 1, 1.5, 2, and 2.5 can accurately fit the seawater conductivity-salinity curve. The temperature and pressure correction terms consider the influence of environmental factors on conductivity measurement, enabling high-precision salinity calculation in various marine environments.

[0108] For the data quality assessment function in step S06, its main calculation process includes:

[0109] Calculation of the parameter correlation matrix:

[0110]

[0111] In the formula, R i，j is the correlation coefficient between parameter i and parameter j; is the k-th sample value of parameter i; is the average value of parameter i; n is the number of samples.

[0112] Calculation of the Mahalanobis distance:

[0113]

[0114] In the formula, D M is the Mahalanobis distance; X is the parameter vector (T int , T ext , P, z, C, S); μ is the parameter mean vector; ∑ is the parameter covariance matrix; ∑ -1 is the inverse of the covariance matrix.

[0115] Calculation of the covariance matrix:

[0116]

[0117] Calculation of the time series mutation index:

[0118]

[0119] Wherein, I anomaly is the mutation index; X t is the parameter value at the current moment; is the parameter value predicted based on a sliding window; σ window is the standard deviation of the parameter within the window.

[0120] Calculation of the sliding window prediction value:

[0121]

[0122] Wherein, w is the window size, and the typical value is 11 - 21.

[0123] The data quality assessment equation system combines statistical and time series analysis methods. The correlation matrix reflects the physical associations between ocean parameters. The Mahalanobis distance can effectively identify outliers in a multi-dimensional space, and the time series mutation index focuses on capturing discontinuous changes in ocean parameters. The advantage of these equations lies in not relying on a single parameter threshold but considering the mutual relationships between parameters, improving the accuracy and robustness of anomaly detection.

[0124] For the dynamic optimization of the temperature difference compensation parameter in step S07, its calculation process is as follows:

[0125] Recursive least squares update formula:

[0126]

[0127] Wherein, θ t is the parameter vector at time t, including the temperature difference compensation polynomial coefficient a i ; K t is the Kalman gain; y t is the actual observed value (temperature deviation); x t is the input vector (1, (T int - T ref ), (T int - T ref ) 2 , (T int - T ref ) 3 ).

[0128] Calculation of the Kalman gain:

[0129]

[0130] Wherein, P t is the parameter covariance matrix, and the initial value is usually set as the identity matrix multiplied by a large constant (such as 100).

[0131] Covariance matrix update:

[0132]

[0133] Weighted objective function:

[0134]

[0135] In the formula, w i is the data point weight, related to the data quality score; λ is the regularization parameter, controlling the deviation degree between the new parameter and the previous parameter, and the typical value is 0.1 - 0.3; θ prev is the previous parameter vector.

[0136] Parameter smooth transition:

[0137] θ applied = αθ new +(1 - α)θ old ;

[0138] In the formula, θ applied is the parameter for actual application; θ new is the newly optimized parameter; θ old is the currently used parameter; α is the smoothing coefficient, and the typical value is 0.15 - 0.3.

[0139] The optimization equation set of the temperature difference compensation parameter is based on the adaptive filtering theory. The recursive least squares method can efficiently update the model parameters when new data arrives without reprocessing the historical data. The weighted objective function takes into account the data quality and reduces the influence of outliers. The regularization term prevents the over - adjustment of parameters and improves the system stability. The smooth transition mechanism avoids the sudden change of parameters and ensures the continuity of the measurement process.

[0140] For the adaptive adjustment of the sampling frequency in step S08, the calculation process is as follows:

[0141] Calculation of the depth change rate:

[0142]

[0143] In the formula, v z is the depth change rate, with the unit of m / s; z(t) is the depth value at time t, with the unit of m; Δt is the time step, and the typical value is 1 s.

[0144] Sampling frequency decision function:

[0145]

[0146] In the formula, f sample is the sampling frequency, with the unit of Hz; v highis the high threshold, with a typical value of 1.5 m / s; v low is the low threshold, with a typical value of 0.5 m / s; t current is the current time; t last is the last frequency switching time; t delay is the minimum switching interval, with a typical value of 3 seconds.

[0147] Power consumption estimation formula:

[0148] P = P base + k f ·f sample ;

[0149] In the formula, P is the system power consumption, with the unit of watt; P base is the basic power consumption, with a typical value of 0.5 watt; k f is the frequency-related coefficient, with a typical value of 0.3 watt / Hz.

[0150] The sampling frequency adjustment equation is designed based on the depth change rate using a piecewise function, and can flexibly adjust the data acquisition density according to the changes in the ocean environment. The hysteresis control mechanism avoids system instability caused by frequent switching. The power consumption estimation formula reflects the linear relationship between the sampling frequency and the system energy consumption, providing a quantitative basis for low-power operation.

[0151] For the calculation of the ocean parameter profile in step S09, the main equations include:

[0152] Seawater density calculation (simplified formula):

[0153] ρ = ρ 0 + A·S + B·T + C·P + D·S·T + E·S·P + F·T·P + G·S·T·P;

[0154] In the formula, ρ is the seawater density, with the unit of kg / m 3 ; ρ 0 is the reference density, with a typical value of 1000 kg / m 3 ; S is the salinity, dimensionless; T is the temperature, with the unit of °C; P is the pressure, with the unit of dbar; A to G are fitting coefficients obtained from experimental data.

[0155] Sound speed calculation hybrid model:

[0156] c = w·c emp + (1 - w)·c NN ;

[0157] In the formula, c is the sound speed, with the unit of m / s; c emp is the sound speed calculated by the empirical formula; c NN is the sound speed predicted by the neural network; w is the weight coefficient, related to the depth.

[0158] Calculation of empirical formula for sound speed:

[0159] c emp = 1449.2 + 4.6T - 0.055T 2 + 0.00029T 3 +(1.34 - 0.01T)(S - 35)+ 0.016P;

[0160] In the formula, T is the temperature, with the unit of °C; S is the salinity, dimensionless; P is the pressure, with the unit of dbar.

[0161] Calculation of weight coefficient:

[0162]

[0163] In the formula, z is the depth, with the unit of meter.

[0164] Cubic spline interpolation formula:

[0165] S i (x)= a i + b i (x - x i )+ c i (x - x i ) 2 + d i (x - xi) 3 , x ∈ [x i , x i+1 ;

[0166] In the formula, S i (x) is the interpolation function within the interval [x i , x i+1 ; a i , b i , c i , d i are the interpolation coefficients, obtained by solving a system of linear equations, satisfying the conditions of continuous values, continuous first derivatives, and continuous second derivatives.

[0167] The system of equations for calculating ocean parameters combines an empirical model and machine learning methods. The density calculation formula takes into account the interaction of temperature, salinity, and pressure, and can accurately reflect the state of seawater. The sound speed mixing model combines the physical interpretability of the empirical formula and the high-precision prediction ability of the neural network. The weight coefficient varies with depth, optimizing the calculation accuracy in different environments. Cubic spline interpolation ensures the smooth transition of standard depth layer data and avoids the discontinuity caused by simple linear interpolation.

[0168] Specifically, the principle of the present invention is as follows: The core principle of the present invention to solve the problem that the temperature difference inside and outside the cabin affects the measurement accuracy lies in constructing a complete dual-temperature measurement and adaptive compensation system. First, the internal temperature sensor and the external temperature sensor are used to measure the temperature inside the cabin and the temperature outside the cabin simultaneously to obtain the basic data of the temperature difference, which is a prerequisite for achieving precise compensation. Different from the traditional single-temperature sensor system, the dual-temperature design can capture the change of the temperature difference inside and outside the cabin in real time, providing a direct basis for the subsequent compensation algorithm.

[0169] Secondly, the adaptive temperature difference compensation function designed in the present invention is the key technical link to solve the problem. Based on the piecewise polynomial fitting method, this function applies different compensation models for different temperature ranges and depth conditions, breaking through the limitations of traditional linear compensation. An adaptive weight allocation mechanism is implemented inside the function, which can adopt a lower weight coefficient for the temperature mutation region to reduce the influence of abnormal data on the compensation effect. At the same time, the self-adjustment of the compensation model is realized through the cumulative error analysis to ensure the stability of the compensation accuracy during the long-term measurement process. This adaptive compensation mechanism can dynamically adjust the compensation coefficient according to the historical temperature difference data, the current temperature change rate, and the environmental conditions, so as to adapt to various complex ocean environments.

[0170] Thirdly, the data quality assessment function and the OceanParamNet model form a double guarantee for data processing and analysis. The data quality assessment function effectively identifies abnormal data points in the measurement process through multi-dimensional parameter correlation analysis and time-series consistency test; while the OceanParamNet model uses a hybrid architecture of variational autoencoder and bidirectional long short-term memory network, which can not only enhance the abnormal detection ability but also predict derivative parameters such as seawater density and sound speed. In particular, the innovative designs such as the sparse attention mechanism, uncertainty estimation module, and physical constraint regularization introduced in the model enable the system to ensure that the prediction results conform to the basic physical laws of oceanography while maintaining the computational efficiency.

[0171] Finally, the present invention adopts a systematic measurement control process for ocean parameters. From system self-check, calibration parameter reading, to parameter measurement, data processing, quality assessment, and result output, a closed-loop control mechanism is formed. In particular, the design of dynamically adjusting the sampling frequency can optimize the system working mode according to the change rate of seawater depth, increase the sampling frequency in the key change region, and decrease the sampling frequency in the stable region, which not only ensures the data quality but also optimizes the system power consumption. The whole system has a rigorous logic, clear levels, accurate functional positioning of each module, and they cooperate with each other to form a complete solution, thus effectively solving the technical problem of unstable measurement accuracy under the condition of temperature difference inside and outside the cabin.

[0172] Next, a specific Embodiment 1 of the present invention is provided, and the specific implementation manners of each step in this Embodiment 1 are described in detail as follows.

[0173] The specific implementation manners of steps S01 - S02 are the same as those described above and will not be elaborated here.

[0174] The specific implementation manner of step S03 is to realize synchronous acquisition and correction of the internal and external cabin temperatures. The temperature sensing device simultaneously acquires the original values of the internal cabin temperature and the external cabin temperature according to a sampling frequency of 2 Hz. During the acquisition process, oversampling technology is adopted, that is, the actual sampling rate is 20 Hz, and it is reduced to 2 Hz for output through mean filtering to improve the signal - to - noise ratio. Subsequently, an adaptive temperature difference compensation function is called to correct the external cabin temperature. This function uses a multi - stage temperature difference compensation algorithm. First, it calculates the statistical characteristics based on the historical temperature difference data sequence (storing the most recent 3600 data points), including the average temperature difference value, standard deviation, and change trend. Then, according to the current temperature change rate (calculated by the temperature change amount within a 10 - second window), the compensation coefficient is dynamically adjusted. The formula for correcting the external cabin temperature is as follows:

[0175] T ext_corrected =T ext_raw +ΔT comp ;

[0176] In the formula, T ext_corrected is the corrected external cabin temperature value, with the unit of °C; T ext_raw is the original temperature value measured by the external cabin temperature sensor, with the unit of °C; ΔT comp is the temperature difference compensation value, with the unit of °C.

[0177] The calculation of the temperature difference compensation value adopts a piece - wise polynomial model:

[0178]

[0179] In the formula, a i is the coefficient of the temperature difference compensation polynomial, usually n is taken as 3; T int is the internal cabin temperature, with the unit of °C; T ref is the reference temperature, with the default value of 20 °C; is the adjustment function related to depth and temperature change rate; d is the current depth, with the unit of meters; is the temperature change rate, with the unit of °C / minute.

[0180] The adjustment function is specifically expressed as:

[0181]

[0182] In the formula, k d is the depth influence coefficient, with the default value of 0.02; k T is the temperature change rate influence coefficient, with the default value of 0.15.

[0183] When the temperature change rate is greater than 0.1 °C per minute, the fast response mode is adopted and the compensation coefficient is increased by 15%; when the temperature change rate is less than 0.02 °C per minute, the steady state mode is adopted and the compensation coefficient is restored to the standard value. Finally, the original temperature data and the corrected temperature data are stored in the data storage device in the form of a time series. This step improves the absolute accuracy of water temperature measurement by precisely compensating for the influence of the temperature difference between inside and outside the cabin.

[0184] The specific implementation of step S04 is to achieve precise processing of pressure measurement and depth conversion. The pressure-depth sensing device collects seawater pressure data at a frequency of 1 Hz and corrects it using the pressure calibration coefficient read in step S02. The correction formula is as follows:

[0185]

[0186] In the formula, P corrected is the corrected pressure value, with the unit of dbar; P raw is the original pressure value measured by the pressure sensor, with the unit of dbar; b i is the pressure calibration polynomial coefficient, which is obtained by calibration with a pressure standard device.

[0187] The corrected pressure value, combined with the current latitude position information, is used to perform depth conversion using the international seawater equation of state:

[0188]

[0189] In the formula, z is the depth, with the unit of meters; g(φ) is the local acceleration of gravity, which is related to the latitude φ; ρ(T, S, P corrected ) is the seawater density, with the unit of kg / m 3 ; T is the temperature, with the unit of °C; S is the salinity, with the unit of PSU; p is the pressure integration variable, with the unit of dbar.

[0190] The formula for calculating the local acceleration of gravity:

[0191] g(φ) = 9.780318 · (1 + 5.2788×10 -3 sin 2 φ + 2.36×10 -5 sin 4 φ);

[0192] Due to the complexity of the integral calculation, an approximate formula is usually used in practical applications:

[0193]

[0194] In the formula, γ is the depth conversion coefficient, and the typical value is 0.99.

[0195] In addition, a pressure hysteresis correction algorithm is applied to compensate for the hysteresis effect of the sensor:

[0196]

[0197] Where P final is the final corrected pressure value; δ h is the hysteresis coefficient, and the typical value range is 0.0005 - 0.002; is the pressure change direction function, which is 1 for rising and -1 for falling. Finally, the original pressure data and the calculated depth data are stored in the data storage device. This step ensures the accuracy and consistency of depth measurement under different sea area conditions.

[0198] The specific implementation of step S05 is to achieve high-precision processing of conductivity measurement and salinity calculation. The conductivity measurement device uses the four-electrode method to collect seawater conductivity data at a frequency of 1 Hz. The four-electrode method effectively eliminates the influence of electrode polarization effect and contact impedance by separating the excitation electrode and the detection electrode. The obtained original conductivity data is corrected using the conductivity calibration coefficient read in step S02, and the correction formula is as follows:

[0199] C corrected = C raw ·(1 + α·(T - T ref ) + β·(T - T ref ) 2 );

[0200] Where C corrected is the corrected conductivity, with the unit of mS / cm; C raw is the original measured conductivity, with the unit of mS / cm; α is the first-order temperature coefficient, and the typical value is 0.02 / °C; β is the second-order temperature coefficient, and the typical value is 0.0005 / °C 2 ; T is the current temperature, with the unit of °C; T ref is the reference temperature, and the typical value is 15°C.

[0201] Conductivity ratio calculation:

[0202]

[0203] Where R is the conductivity ratio, dimensionless; C KCl is the conductivity of the standard KCl solution at the same temperature, with the unit of mS / cm.

[0204] Salinity calculation uses the practical salinity standard formula:

[0205] S = a 0 + a 1 R 1 / 2 + a 2 R + a3 R 3 / 2 +a 4 R 2 +a 5 R 5 / 2 +ΔS(T, P);

[0206] where S is the practical salinity, dimensionless; a 0 to a 5 are standard coefficients, with values of 0.0080, -0.1692, 25.3851, 14.0941, -7.0261, 2.7081 respectively; ΔS(T, P) is the temperature and pressure correction term.

[0207] Calculation of the temperature and pressure correction term:

[0208]

[0209] where b 0 to b 5 are pressure correction coefficients; k is the temperature influence coefficient, with a typical value of 0.0162; m is the temperature normalization coefficient, with a typical value of 0.0116; P is the pressure, in dbar; T is the temperature, in °C. Finally, the original conductivity data and the calculated salinity data are stored in the data storage device. This step ensures the accuracy and reliability of salinity measurement in a complex marine environment.

[0210] The specific implementation of step S06 is to achieve a comprehensive assessment of multi-parameter data quality. This step calls the data quality assessment function and the OceanParamNet model to analyze the in-cabin temperature, out-cabin temperature, pressure, depth, conductivity, and salinity data obtained in steps S03 to S05. The data quality assessment function first calculates the parameter correlation matrix:

[0211]

[0212] where R i，j is the correlation coefficient between parameter i and parameter j; is the k-th sample value of parameter i; is the average value of parameter i; n is the number of samples.

[0213] Subsequently, the normal correlation range is determined based on historical statistical data. For example, the correlation coefficient between temperature and salinity should usually be in the range of -0.7 to 0.3 in a specific sea area. Then, the Mahalanobis distance algorithm is applied to detect outliers in the multi-dimensional parameter space:

[0214]

[0215] where D M is the Mahalanobis distance; X is the parameter vector (T int , Text , P, z, C, S); μ is the parameter mean vector; ∑ is the parameter covariance matrix; ∑ -1 is the inverse of the covariance matrix. When the distance exceeds 3.5 standard deviations, it is marked as an anomaly.

[0216] At the same time, the sliding window technique is applied to analyze the temporal coherence of the parameters and calculate the mutation index:

[0217]

[0218] In the formula, I anomaly is the mutation index; X t is the parameter value at the current moment; is the parameter value predicted based on the sliding window; σ window is the standard deviation of the parameters within the window. When the mutation index exceeds a preset threshold (usually 2.8), it is marked as a temporal anomaly.

[0219] The OceanParamNet model reconstructs the input data through its variational autoencoder branch, calculates the reconstruction error, and further confirms the abnormal state when the error exceeds the dynamic threshold. Abnormal data is classified into two categories: correctable and uncorrectable according to the severity. Correctable data is replaced by interpolation or model prediction, and uncorrectable data is marked for deletion. All data quality assessment results are stored in the data storage device in a marked form. This step improves the reliability and scientific value of the measurement data through a method combining multi-dimensional analysis and machine learning.

[0220] The specific implementation of step S07 is to realize the dynamic optimization and adjustment of the temperature difference compensation parameters. This step is based on the updated value of the compensation coefficient output by the adaptive temperature difference compensation function in step S03 and the abnormal data correction suggestions provided by the data quality assessment function in step S06, and evaluates the impact of the temperature change inside and outside the cabin on the measurement results in real time. The specific implementation is to construct a temperature difference impact assessment model, which uses the recursive least squares method to update the temperature difference compensation parameters:

[0221]

[0222] In the formula, θ t is the parameter vector at time t, including the temperature difference compensation polynomial coefficient a i ; K t is the Kalman gain; y t is the actual observed value (temperature deviation); x t is the input vector (1, (T int -T ref ), (T int -T ref ) 2 , (T int -T ref ) 3)。

[0223] Kalman gain calculation:

[0224]

[0225] Covariance matrix update:

[0226]

[0227] When the rate of change of the temperature inside the cabin is detected to exceed 0.2 °C / minute, trigger the process of re-evaluating the compensation parameters by optimizing the weighted objective function:

[0228]

[0229] where w i is the weight of the data point, related to the data quality score; λ is the regularization parameter, controlling the deviation degree between the new parameter and the previous parameter, and the typical value is 0.1 - 0.3; θ prev is the previous parameter vector.

[0230] The newly obtained compensation parameters after optimization are applied to the actual system through smooth transition:

[0231] θ applied = αθ new +(1 - α)θ old ;

[0232] where θ applied is the parameter for actual application; θ new is the newly optimized parameter; θ old is the currently used parameter; α is the smoothing coefficient, and the typical value is 0.15 - 0.3.

[0233] To prevent violent fluctuations of the parameters, set the upper limit of the single adjustment amplitude, not exceeding ±15% of the previous parameter. The optimized parameter values are recorded in the data storage device together with the timestamps. This step ensures the stability of the measurement accuracy in an environment with rapid temperature changes through an adaptive adjustment mechanism.

[0234] The specific implementation of step S08 is to achieve the adaptive adjustment control of the sampling frequency. This step dynamically adjusts the sampling frequencies of the pressure-depth sensing device and the conductivity measurement device according to the rate of change of the seawater depth measured in step S04. The rate of change of the depth is calculated by the central difference method:

[0235]

[0236] where v z is the rate of change of the depth, with the unit of m / s; z(t) is the depth value at time t, with the unit of m; Δt is the time step, and the typical value is 1 second.

[0237] Sampling frequency decision function:

[0238]

[0239] where f sample is the sampling frequency, with the unit of Hertz; v high is the high threshold, with a typical value of 1.5 m / s; v low is the low threshold, with a typical value of 0.5 m / s; t current is the current time; t last is the last frequency switching time; t delay is the minimum switching interval, with a typical value of 3 seconds.

[0240] To prevent the sampling frequency from switching frequently, a hysteresis control mechanism is introduced, that is, the frequency adjustment is triggered only when the change rate continuously exceeds the threshold for more than 3 seconds. At the same time, the system power consumption management module adjusts the power distribution according to the current sampling frequency, and the power consumption estimation formula:

[0241] P = P base + k f · f sample ;

[0242] where P is the system power consumption, with the unit of Watt; P base is the base power consumption, with a typical value of 0.5 Watt; k f is the frequency-related coefficient, with a typical value of 0.3 Watt / Hertz.

[0243] The adjusted sampling frequency is applied to the data acquisition process of steps S03 to S05 in real time. This step optimizes the system power consumption while ensuring the data density in the key area by intelligently adjusting the sampling strategy, and extends the battery life of the device.

[0244] The specific implementation of step S09 is to realize the comprehensive calculation and construction of the ocean parameter profile. This step calls the deep fully connected network branch of the OceanParamNet model, and calculates the seawater density, sound speed and other ocean physical parameters based on the seawater temperature, salinity and depth data obtained in steps S03 to S05. The seawater density calculation uses the simplified formula:

[0245] ρ = ρ 0 + A · S + B · T + C · P + D · S · T + E · S · P + F · T · P + G · S · T · P;

[0246] where ρ is the seawater density, with the unit of kg / m 3 ; ρ 0 is the reference density, with a typical value of 1000 kg / m 3; S is salinity, dimensionless; T is temperature, in °C; P is pressure, in dbar; A to G are fitting coefficients.

[0247] The sound speed calculation uses a hybrid model:

[0248] c = w·c emp +(1 - w)·c NN ;

[0249] In the formula, c is the sound speed, in m / s; c emp is the sound speed calculated by the empirical formula; c NN is the sound speed predicted by the neural network; w is the weight coefficient, related to depth.

[0250] Empirical formula for sound speed calculation:

[0251] c emp = 1449.2 + 4.6T - 0.055T 2 + 0.00029T 3 +(1.34 - 0.01T)(S - 35)+ 0.016P;

[0252] Variation of the weight coefficient with depth:

[0253]

[0254] The calculation results include the parameter values of the standard depth layer, and cubic spline interpolation is used:

[0255] S i (x)= a i + b i (x - x i )+ c i (x - x i ) 2 + d i (x - x i ) 3 , x ∈ [x i , x i+1 ;

[0256] The finally constructed ocean parameter profile is stored in the data storage device in a structured format. This step provides a comprehensive analysis of ocean parameters through a method combining deep learning and physical models, providing high-quality data support for oceanographic research.

[0257] The specific implementation of step S10 is the same as the foregoing, and will not be elaborated here.

[0258] To better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: First, during the power-on self-check phase of the system, the researchers recorded the initial states of each sensor. The response time of the internal temperature sensor was 42 milliseconds, and the signal fluctuation was ±0.004 °C; the response time of the external temperature sensor was 75 milliseconds, and the signal fluctuation was ±0.002 °C; the zero offset of the pressure sensor was 0.03% of the full scale; the impedance value between the four electrodes of the conductivity measuring device was 4.2 kiloohms. All sensors were within the normal range, and the self-check result was good.

[0259] The system read calibration parameters from the non-volatile storage area, including temperature calibration coefficients, pressure calibration coefficients, and conductivity calibration coefficients. The core calibration parameters are shown in Table 1:

[0260] Table 1 System Calibration Parameter Table

[0261] Parameter type Parameter value Valid range <![CDATA[Temperature calibration coefficient a 0 > 0.0023 ±0.0005 <![CDATA[Temperature calibration coefficient a 1 > 0.9973 ±0.0100 <![CDATA[Temperature calibration coefficient a 2 > 0.0002 ±0.0001 <![CDATA[Temperature calibration coefficient a 3 > 0.00005 ±0.00002 <![CDATA[Pressure calibration coefficient b 0 > 0.0347 ±0.0050 <![CDATA[Pressure calibration coefficient b 1 > 0.9986 ±0.0100 <![CDATA[Pressure calibration coefficient b 2 > 0.00003 ±0.00001 <![CDATA[Pressure calibration coefficient b 3 > -0.000001 ±0.0000005 First-order temperature coefficient α of conductivity 0.0195 ±0.0020 Second-order temperature coefficient β of conductivity 0.00048 ±0.00005

[0262] During actual measurement, the system synchronously collected the temperature inside the cabin and the temperature outside the cabin at a frequency of 2 Hz. The researchers recorded the data at a certain moment during the dive. The temperature inside the cabin was 18.475 °C, the original reading of the temperature outside the cabin was 5.873 °C, the depth was 325 meters, and the temperature change rate was 0.085 °C / minute. The system applied the adaptive temperature difference compensation function for correction, and the temperature difference compensation parameter k d was set to 0.02, and k T was set to 0.15. According to the piecewise polynomial model, the calculated temperature difference compensation value ΔT comp was 0.063 °C, and the corrected temperature outside the cabin was 5.936 °C.

[0263] At the same time, the pressure-depth sensing device collected seawater pressure data at a frequency of 1 Hz. At this depth point, the original pressure reading was 327.85 dbar, and the corrected pressure after applying the third-order polynomial calibration was 328.12 dbar. Considering that the latitude of the measurement sea area was 21.35°N, the system calculated the local acceleration of gravity to be 9.788 m / s 2 , and the accurate depth was obtained as 325.43 meters by applying the international seawater equation of state.

[0264] Table 2 shows the comparison of the original pressure and the calibrated depth data of the system at different depth points:

[0265] Table 2 Pressure and Depth Calibration Comparison Table

[0266] Original pressure (dbar) Corrected pressure (dbar) Converted depth (m) Hysteresis correction (dbar) 101.38 101.52 100.57 +0.05 202.93 203.14 201.28 +0.10 327.85 328.12 325.43 +0.16 450.72 451.08 447.36 +0.23 582.45 582.89 578.11 +0.29

[0267] The conductivity measurement device uses the four-electrode method to collect seawater conductivity data at a frequency of 1 Hz. At a depth of 325.43 meters, the original conductivity reading was 42.357 mS / cm, and the temperature was 5.936 °C. The conductivity after applying temperature correction was 43.125 mS / cm, and the calculated conductivity ratio R was 1.0963. Using the practical salinity standard formula, the system calculated the salinity at this point to be 34.527 PSU.

[0268] The researchers recorded the conductivity and salinity calculation results of the system at different depths in this depth profile, as shown in Table 3:

[0269] Table 3 Conductivity and Salinity Calculation Results

[0270]

[0271]

[0272] In the data quality assessment section, the system calculated the parameter correlation matrix, as shown in Table 4:

[0273] Table 4 Parameter Correlation Matrix

[0274]

[0275] The system used the Mahalanobis distance algorithm to detect outliers. Among the 500 data points collected in this profile, 7 potential outliers were identified, with Mahalanobis distances exceeding 3.5 standard deviations. Through time series mutation index analysis, 5 of them were confirmed as real outliers, and the sliding window technique (window size of 15 data points) was applied to correct these outliers.

[0276] To further optimize the temperature difference compensation parameters, the system applied the recursive least squares method to update the parameters in real time. By analyzing the correlation between temperature difference changes and measurement errors in historical data, new compensation coefficients were calculated. The comparison of temperature difference compensation parameters before and after optimization is shown in Table 5:

[0277] Table 5 Comparison of Temperature Difference Compensation Parameter Optimization

[0278] Parameter Initial value Optimized value Rate of change (%) <![CDATA[a 0 > 0.0023 0.0025 +8.7 <![CDATA[a 1 > 0.0058 0.0063 +8.6 <![CDATA[a 2 > 0.0002 0.0002 0.0 <![CDATA[a 3 > 0.00005 0.00006 +20.0 <![CDATA[k d > 0.0200 0.0210 +5.0 <![CDATA[k T > 0.1500 0.1575 +5.0

[0279] The system adaptively adjusts the sampling frequency according to the depth change rate. In this measurement, the depth change rate and sampling frequency adjustment in different diving stages were recorded, as shown in Table 6:

[0280] Table 6 Depth Change Rate and Sampling Frequency Adjustment

[0281] Time period (min:sec) Average rate of change (m / s) Sampling frequency (Hz) Power consumption estimate (W) 00:00-05:30 1.75 2.0 1.10 05:30-12:45 1.32 1.0 0.80 12:45-18:20 0.85 1.0 0.80 18:20-25:10 0.42 0.5 0.65 25:10-32:30 0.13 0.5 0.65

[0282] In the calculation of ocean parameter profiles, the system used a hybrid model to calculate seawater density and sound speed. At a depth of 325.43 meters, the temperature was 5.936 °C, the salinity was 34.527 PSU, and the calculated seawater density was 1027.18 kg / m 3 , and the sound speed was 1493.56 m / s. The calculation results of the ocean parameters at each standard depth layer of the system are shown in Table 7:

[0283] Table 7 Calculation Results of Ocean Parameters at Standard Depth Layers

[0284] The system regularly uploads the collected and calculated data to the host computer through the communication device, following the hierarchical transmission strategy. During the entire observation process, the system uploaded a total of 3250 data packets, including 1286 key parameter data packets, 643 derived parameter data packets, and 1321 system status information data packets. The data transmission success rate was 99.7%, and only 10 data packets were retransmitted due to communication interference.

[0285] Traditional temperature-salinity-depth measurement systems usually use a single temperature sensor design, ignoring the impact of temperature differences inside and outside the instrument on measurement accuracy. This design will produce obvious errors in environments with large temperature changes, especially when conducting profile measurements in thermoclines or deep-sea environments. Traditional systems use a fixed sampling frequency and a simple linear calibration model, which cannot adapt to the changes in complex ocean environments, and the data quality assessment mainly relies on single-parameter threshold judgment, lacking the ability of multi-dimensional correlation analysis.

[0286] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Tables 8 and 9 below.

[0287] Table 8 Variable Explanation Table (Part 1)

[0288] The following are the explanations of all variables, subscripts, and constants involved in the full text:

[0289]

[0290]

[0291] Table 9 Variable Explanation Table (Part 2)

[0292]

[0293] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

Claims

1. A dual-temperature compensation salt depth instrument inside and outside the cabin, comprising a control chip, a temperature sensor, a pressure depth sensor, a conductivity measurement device, a data storage device and a communication device, wherein the control chip is electrically connected to the temperature sensor, the pressure depth sensor, the conductivity measurement device, the data storage device and the communication device, respectively, and an ocean parameter measurement control module is provided in the control chip; characterized in that: The temperature sensing device includes an internal temperature sensor and an external temperature sensor, which are used to simultaneously measure the temperature inside the cabin and the temperature outside the cabin and calculate the temperature difference compensation value; the ocean parameter measurement control module is used to perform self-test, read system calibration parameters, collect temperature data, collect pressure data, collect conductivity data, data quality assessment, dynamically adjust temperature difference compensation parameters, adaptively adjust sampling frequency, calculate seawater parameters and upload measurement data steps; the ocean parameter measurement control module calls the data quality assessment function and the OceanParamNet model in the data quality assessment step to comprehensively analyze the temperature inside the cabin, the temperature outside the cabin, pressure, depth, conductivity and salinity data.

2. The dual-temperature compensation salt depth meter for inside and outside the cabin according to claim 1 is characterized in that: The measurement accuracy of the internal temperature sensor is ±0.01°C, the measurement accuracy of the external temperature sensor is ±0.005°C, and the sampling frequency of the temperature sensing device is 2 times per second.

3. The dual-temperature compensation salt depth meter for inside and outside the cabin according to claim 2 is characterized in that: The measuring range of the pressure depth sensor device is 0-1000 meters, the measuring accuracy is ±0.1% of full scale, and the sampling frequency of the pressure depth sensor device is once per second.

4. The dual-temperature compensation salt depth meter for inside and outside the cabin according to claim 3 is characterized in that: The conductivity measuring device adopts a four-electrode method for measurement, with a measurement range of 0-70 mS / cm, a measurement accuracy of ±0.005 mS / cm, and a sampling frequency of once per second.

5. The dual-temperature compensation salt depth meter for inside and outside the cabin according to claim 4 is characterized in that: The input of the adaptive temperature difference compensation function includes the cabin temperature value measured by the temperature sensing device, the original value of the cabin temperature outside the cabin measured by the temperature sensing device, the historical temperature difference data sequence stored in the data storage device, the instrument depth value measured by the pressure depth sensing device and the temperature change rate calculated by the temperature sensing device; the output of the adaptive temperature difference compensation function is the cabin temperature correction value and the compensation coefficient update value after precise compensation.

6. The dual-temperature compensation salt depth meter for inside and outside the cabin according to claim 5 is characterized in that: The adaptive temperature difference compensation function adopts a piecewise polynomial fitting method, applies different compensation models to different temperature ranges and depth conditions, and dynamically adjusts compensation parameters according to the statistical characteristics of the historical temperature difference data.

7. The dual-temperature compensation salt depth meter for inside and outside the cabin according to claim 6 is characterized in that: The adaptive temperature difference compensation function implements an adaptive weight distribution mechanism internally, adopts a lower weight coefficient for the temperature mutation area, reduces the impact of abnormal data on the compensation effect, and realizes self-adjustment of the compensation model through cumulative error analysis, ensuring the stability and reliability of compensation accuracy during long-term measurement.

8. The dual-temperature compensation salt depth meter for inside and outside the cabin according to claim 7 is characterized in that: The data quality assessment function identifies abnormal data points through multi-dimensional parameter correlation analysis and time series consistency check. The OceanParamNet model further enhances the anomaly detection capability through its variational autoencoder branch, and marks or removes abnormal data. The input of the data quality assessment function includes the cabin temperature data measured by the temperature sensor device, the cabin temperature data measured by the temperature sensor device, the pressure data measured by the pressure depth sensor device, the depth data calculated by the pressure depth sensor device, the conductivity data measured by the conductivity measuring device, and the salinity data calculated according to the practical salinity standard formula. The output of the data quality assessment function is the data quality marking result.

9. The dual-temperature compensation salt depth meter for inside and outside the cabin according to claim 8 is characterized in that: The OceanParamNet model is a deep learning model for ocean parameter association. Its specific structure is a hybrid architecture based on a variational autoencoder and a bidirectional long short-term memory network. The front end uses a multi-layer perceptron to extract features from the original measurement data, the middle layer is composed of a bidirectional long short-term memory network, and the back end is divided into two branches. One branch reconstructs input data for anomaly detection through a variational autoencoder, and the other branch predicts derived parameters of seawater density and sound speed through a deep fully connected network. The OceanParamNet model introduces a sparse attention mechanism, and the attention window size of the sparse attention mechanism is dynamically adjusted according to the sampling frequency of the temperature sensor device and the seawater depth change rate measured by the pressure depth sensor device, ensuring that more computing resources are allocated in key change areas.

10. The dual-temperature compensation salt depth meter for inside and outside the cabin according to claim 9 is characterized in that: The data quality assessment function also includes a gradient analysis module, which identifies physically impossible mutations by checking the parameter change rate and designs different marking mechanisms for different types of anomalies, providing an accurate quality control basis for subsequent data processing.

Images