Multi-channel signal acquisition and analysis system for safe navigation of ship

Abstract

The invention provides a multi-channel signal acquisition and analysis system for safe navigation of a ship, and relates to the technical field of signal acquisition, and the system comprises a signal acquisition layer which comprises signal acquisition equipment and a signal conditioning module which comprises a programmable filter, a gain amplifier and an analog-to-digital converter; the signal processing layer comprises a logic control module, and the logic control module adopts an FPGA (Field Programmable Gate Array) and a reconfigurable data flow pipeline to carry out synchronous signal processing; the data analysis layer comprises an edge processing unit, an analysis algorithm library and a real-time decision unit; and the control and communication layer comprises a system control unit, a data transmission module and a storage management unit, the system control unit performs working mode management according to an instruction of the upper computer and is provided with a resource scheduler and a power consumption manager, and the data transmission module is provided with a multiple access transmission controller, a data compression unit and an encryption transmission module. According to the invention, comprehensive integration from signal acquisition to data analysis is realized, and the safe navigation capability of the ship is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of signal acquisition, and in particular to a multi-channel signal acquisition and analysis system for safe navigation of ships. Background Art

[0002] Maritime safety is a top priority for the global shipping industry. With the continuous expansion of ship size and the increasing complexity of routes, the requirements for ship navigation systems are also getting higher and higher. Traditional ship navigation mainly relies on single information sources such as GPS, radar and AIS, which may not provide sufficiently comprehensive and accurate navigation information in some cases. Therefore, multi-source data fusion technology has been widely used in the field of ship navigation. By integrating multiple sensor data, it can more comprehensively perceive the surrounding environment of the ship and improve navigation accuracy and reliability.

[0003] In recent years, with the rapid development of artificial intelligence and big data technology, intelligent ship navigation systems have gradually become a research hotspot. Such systems can not only process and analyze massive amounts of data in real time, but also make intelligent decisions based on complex sea and weather conditions. The introduction of edge computing technology enables data processing to be closer to the data source, greatly reducing data transmission delays and improving the real-time performance of the system. At the same time, the application of deep learning algorithms in target recognition, route planning, etc. has further improved the intelligence level of the navigation system.

[0004] At present, ship navigation systems are developing towards a more integrated, intelligent and autonomous direction. With the increasing awareness of marine environmental protection, low-power and environmentally friendly navigation systems have also become a research focus. However, the real-time fusion and processing of multi-source heterogeneous data is still a difficult problem, especially in the complex and changeable marine environment. How to effectively integrate data from different sensors and conduct real-time analysis and decision-making under limited computing resources still requires further technological breakthroughs. Summary of the invention

[0005] In view of this, the present invention proposes a multi-channel signal acquisition and analysis system for safe navigation of ships. By integrating data from multiple sensors, real-time fusion processing of multi-source heterogeneous information is realized, and artificial intelligence and edge computing technologies are used for intelligent analysis and decision-making to improve navigation accuracy and reliability, enhance adaptability and autonomy in complex marine environments, and at the same time, by adopting reconfigurable computing technology and intelligent algorithm selection mechanism, the system can adapt to different navigation requirements and environmental changes.

[0006] The technical solution of the present invention is implemented as follows: The present invention provides a multi-channel signal acquisition and analysis system for safe navigation of ships, comprising:

[0007] The signal acquisition layer includes signal acquisition equipment and signal conditioning modules. The signal acquisition equipment includes GPS, IMU, ultrasonic sensor, barometer, infrared thermal imager, radar system, underwater sonar system, AIS receiver, and the signal conditioning module includes programmable filter, gain amplifier and analog-to-digital converter.

[0008] The signal processing layer includes a logic control module, which uses FPGA and a reconfigurable data flow pipeline to perform synchronous signal processing;

[0009] The data analysis layer includes an edge processing unit, an analysis algorithm library, and a real-time decision unit. The edge processing unit uses a deep learning accelerator and a neural network inference engine for data analysis. The analysis algorithm library contains various algorithms and is equipped with an algorithm selection mechanism. The real-time decision unit generates real-time decisions based on the data analysis results.

[0010] The control and communication layer includes a system control unit, a data transmission module and a storage management unit. The system control unit manages the working mode according to the instructions of the upper computer and is configured with a resource scheduler and a power consumption manager. The data transmission module is configured with a multi-address transmission controller, a data compression unit and an encryption transmission module. The storage management unit uses a high-speed cache and a large-capacity storage interface to store data.

[0011] On the basis of the above scheme, preferably, the signal conditioning module further includes a preamplifier, an isolation circuit, a temperature compensation system and a power supply module, and the programmable filter is a bandpass filter with adjustable bandwidth, wherein:

[0012] The preamplifier integrates AGC function to adjust the gain according to the input signal strength; the temperature compensation system compensates for the impact of temperature changes on the signal; the power module provides multi-channel power supply.

[0013] On the basis of the above scheme, preferably, the logic control module integrates DSP and embedded microprocessor, and the logic control module has built-in adaptive filtering algorithm, fuzzy control algorithm, PID control algorithm, temperature compensation algorithm, wherein:

[0014] The adaptive filtering algorithm is configured in the DSP to control the programmable filters and adjust the parameters of each filter;

[0015] The fuzzy control algorithm is configured in the embedded microprocessor for AGC control of the preamplifier;

[0016] The PID control algorithm is configured in the FPGA and used to control the gain amplifier;

[0017] The temperature compensation algorithm is configured in the embedded microprocessor and is used to optimize the temperature compensation model in the temperature compensation system.

[0018] Based on the above solution, preferably, the expression of the temperature compensation model is:

[0019] S c =S m ×(1+α(TT r )+β(TT r ) 2 )+γ

[0020] In the formula, S c is the signal value after temperature compensation, S m is the measured original signal value, T is the current temperature, T r is the reference temperature, α is the first-order temperature coefficient, β is the second-order temperature coefficient, and γ is the offset correction factor.

[0021] Based on the above solution, preferably, the temperature compensation algorithm is:

[0022] Step 1: Set the initial parameters α, β, γ; set the learning rate η to control the amplitude of parameter update in each iteration; set the number of iterations N max ;

[0023] Step 2: Collect signal acquisition device data and corresponding true values ​​at different temperatures;

[0024] Step 3: Set the number of iterations = 1;

[0025] Step 4. Select a data point x i ;

[0026] Step 5: Calculate the temperature compensation model for the data point x i The predicted value of

[0027]

[0028] Step 6: According to the predicted value and data point x i The true value of Calculation error E(x i ):

[0029]

[0030] Step 7: Update the parameters α, β, and γ according to the error:

[0031] α=α-η×v α

[0032] β=β-η×v β

[0033] γ=γ-η×v γ

[0034]

[0035]

[0036] In the formula, v α 、v β 、v γ is the momentum term, μ is the momentum coefficient, represents partial derivative;

[0037] Step 8: Repeat steps 4 to 7 to traverse all data points;

[0038] Step 9: Determine whether the number of iterations reaches N max If yes, then output the trained temperature compensation model; if no, set the number of iterations +1 and return to step 4;

[0039] Step 10: Use the trained temperature compensation model to perform real-time temperature compensation;

[0040] Step 11: Periodically collect new data and use the new data to fine-tune the temperature compensation model.

[0041] Based on the above scheme, preferably, the algorithm selection mechanism is as follows:

[0042] Scan the input data to extract key meta-features, including data type, scale, dimension, and sparsity, and classify the tasks according to the preset goals;

[0043] Based on the task type and key meta-features of the data, a predefined heuristic rule is used to perform preliminary screening to obtain the first set of candidate algorithms;

[0044] Using historical performance data and meta-learning models, combined with current resources, the first candidate algorithm set is screened again to obtain a second candidate algorithm set;

[0045] The comprehensive score of each algorithm in the second candidate algorithm set is calculated according to the scoring formula, and the algorithm with the highest comprehensive score is selected as the optimal algorithm.

[0046] Based on the above solution, preferably, the execution process of the predefined heuristic rules is as follows:

[0047] Design predefined heuristic rules, including rules based on task type and rules based on data features;

[0048] Initialize an empty set of candidate algorithms;

[0049] According to the task type, add the corresponding algorithm to the candidate algorithm set;

[0050] Apply the rules one by one based on the data characteristics:

[0051] If the data scale is large, keep the algorithm that can process big data;

[0052] If the data is high-dimensional, add a dimensionality reduction algorithm;

[0053] If the data is highly nonlinear, the nonlinear algorithm is retained;

[0054] If the data categories are unbalanced, add algorithms to handle unbalanced data;

[0055] If the data is sparse, keep the algorithm that handles sparse data;

[0056] Assign preliminary priorities to candidate algorithms based on the degree of rule matching;

[0057] Set the maximum candidate algorithm number threshold N and the matching degree threshold p, sort the candidate algorithms according to the matching degree from large to small, first filter the candidate algorithms according to the matching degree threshold p, if the number of filtered candidate algorithms does not exceed the maximum candidate algorithm number threshold N, then use the filtered candidate algorithms as the first candidate algorithm set; if the number of filtered candidate algorithms exceeds the maximum candidate algorithm number threshold N, then retain the top N algorithms as the first candidate algorithm set according to the preliminary priority.

[0058] Based on the above scheme, preferably, the scoring formula is as follows:

[0059] F=G(P,E,R,A,I,S)+D(M,L)

[0060]

[0061] D(M,L)=ω m ×M+ω l ×L

[0062] Where F is the comprehensive score, G(P,E,R,A,I,S) is the main scoring function, P is the historical performance score, E is the computational efficiency score, R is the resource usage score, A is the adaptability score, I is the interpretability score, S is the stability score, δ,ε,∈,σ,ξ,κ are dynamically adjusted parameters, D(M,L) is the difference adjustment function, M is the complexity control factor, L is the learning adaptability, ω m ,ω l is an adjustable trade-off parameter.

[0063] Based on the above scheme, preferably, the preamplifier adopts AD8367, AD8367 has a built-in AGC function, the isolation circuit adopts ADuM1401, the temperature compensation system adopts ADT7420, the programmable filter adopts LTC1068, the gain amplifier adopts AD8338, the analog-to-digital converter adopts ADS1258, and the power module adopts ADP5052.

[0064] On the basis of the above scheme, preferably, the logic control module adopts the control chip of ZYNQ7000, which integrates a dual-core ARM Cortex-A9 processor, programmable logic and hard-core DSP Slices, implements reconfigurable data flow pipeline and PID control in the FPGA part, uses DSP Slices to implement adaptive filtering, uses the dual-core ARM Cortex-A9 processor as an embedded microprocessor, and runs fuzzy control and temperature compensation algorithms on the embedded microprocessor.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] (1) Through multi-layer architecture design, comprehensive integration from signal acquisition to data analysis is achieved, greatly improving the ship's safe navigation capabilities. The system can process multi-source heterogeneous data in real time, perform intelligent analysis and decision-making, and significantly improve navigation accuracy and reliability. At the same time, the system's modular design and reconfigurable characteristics make it highly adaptable and scalable, and can cope with various complex marine environments and navigation requirements;

[0067] (2) The signal conditioning module significantly improves the quality and reliability of the signal by integrating the preamplifier, isolation circuit, temperature compensation system and anti-aliasing filter. The AGC function of the preamplifier can dynamically adjust the gain to ensure that the signal always remains within the optimal amplitude range. The temperature compensation system effectively reduces the impact of temperature changes on the signal and improves the stability of the system under various environmental conditions. The adaptive anti-aliasing filter can automatically adjust the cutoff frequency according to the sampling rate, effectively preventing the occurrence of aliasing and ensuring the integrity and accuracy of the signal;

[0068] (3) The logic control module integrates DSP and embedded microprocessor, and has built-in multiple control algorithms to achieve efficient signal processing and system control. The adaptive filtering algorithm can adjust the filter parameters in real time to adapt to different signal characteristics. The fuzzy control algorithm improves the intelligence level of the preamplifier and anti-aliasing filter. The PID control algorithm ensures the precise control of the gain amplifier. The temperature compensation algorithm further optimizes the temperature compensation effect and improves the stability and accuracy of the system under different temperature environments;

[0069] (4) The temperature compensation model achieves accurate compensation for the effects of temperature changes by introducing first-order and second-order temperature coefficients and offset correction factors. This compensation model can effectively reduce the interference of temperature changes on the signal; and the corresponding adaptive temperature compensation algorithm can automatically adjust the parameters of the temperature compensation model through iterative optimization and momentum update mechanism;

[0070] (5) The algorithm selection mechanism can select the optimal algorithm under different tasks and data characteristics through multi-step screening and comprehensive scoring. This mechanism improves the flexibility and efficiency of the system, enabling it to adapt to various complex data analysis tasks and achieve optimal performance under limited computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0072] Figure 1 is a system architecture diagram of an embodiment of the present invention;

[0073] Figure 2 Schematic diagram of the AGC circuit of the preamplifier according to an embodiment of the present invention;

[0074] Figure 3 is a functional block diagram of an isolation circuit according to an embodiment of the present invention;

[0075] Figure 4 is a functional block diagram of a temperature compensation system according to an embodiment of the present invention;

[0076] Figure 5 It is a schematic diagram of a programmable filter circuit according to an embodiment of the present invention;

[0077] Figure 6 is a circuit diagram of a gain amplifier according to an embodiment of the present invention;

[0078] Figure 7 The AD conversion interface circuit diagram of the analog-to-digital converter according to the embodiment of the present invention;

[0079] Figure 8 A power application circuit diagram of a power module according to an embodiment of the present invention;

[0080] Fig. 9 is a block diagram of the internal structure of a logic control module according to an embodiment of the present invention;

[0081] Fig.10 Flow chart of a temperature compensation algorithm according to an embodiment of the present invention. DETAILED DESCRIPTION

[0082] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0083] like Figure 1 As shown, the present invention provides a multi-channel signal acquisition and analysis system for safe navigation of ships, comprising:

[0084] The signal acquisition layer includes signal acquisition equipment and signal conditioning modules. The signal acquisition equipment includes GPS, IMU, ultrasonic sensor, barometer, infrared thermal imager, radar system, underwater sonar system, AIS receiver, and the signal conditioning module includes programmable filter, gain amplifier and analog-to-digital converter.

[0085] The signal processing layer includes a logic control module, which uses FPGA and a reconfigurable data flow pipeline to perform synchronous signal processing;

[0086] The data analysis layer includes an edge processing unit, an analysis algorithm library, and a real-time decision unit. The edge processing unit uses a deep learning accelerator and a neural network inference engine for data analysis. The analysis algorithm library contains various algorithms and is equipped with an algorithm selection mechanism. The real-time decision unit generates real-time decisions based on the data analysis results.

[0087] The control and communication layer includes a system control unit, a data transmission module and a storage management unit. The system control unit manages the working mode according to the instructions of the upper computer and is configured with a resource scheduler and a power consumption manager. The data transmission module is configured with a multi-address transmission controller, a data compression unit and an encryption transmission module. The storage management unit uses a high-speed cache and a large-capacity storage interface to store data.

[0088] The system execution process of the present invention is as follows: 1) The system starts and each module is initialized; the system control unit receives the host computer instruction and sets the working mode; the resource scheduler allocates system resources, and the power consumption manager optimizes energy use; 2) Multiple signal acquisition devices (GPS, IMU, ultrasound, barometer, infrared thermal imager, radar, sonar, AIS receiver) collect data simultaneously; the signal conditioning module processes the original signal; the analog-to-digital converter converts the analog signal into a digital signal; 3) FPGA receives the digitized signal; a reconfigurable data flow pipeline is used for parallel signal processing; the logic control module coordinates the synchronous processing of different signal sources; 4) The edge processing unit receives the processed data: a. The deep learning accelerator performs preliminary data processing; b. The neural network inference engine Perform feature extraction and preliminary classification; the algorithm selection mechanism selects the most suitable algorithm from the analysis algorithm library; executes the selected data analysis algorithm; the real-time decision unit generates decision recommendations based on the analysis results; 5) The system control unit adjusts the system parameters based on the analysis results and decision recommendations; the data transmission module processes data: a. The multi-access transmission controller manages the data flow; b. The data compression unit compresses the data; c. The encryption transmission module ensures data security; the storage management unit stores key data in the cache and large-capacity storage devices; 6) The system continuously monitors performance and effects; adjusts resource allocation and power consumption management according to real-time conditions; optimizes algorithm selection and data analysis processes; 7) The system continuously executes steps 2)-6), processes new collected data in real time; and reports status and results to the host computer as needed.

[0089] Specifically, in one embodiment of the present invention, the signal acquisition layer is described as follows:

[0090] Among the signal acquisition equipment, GPS provides accurate position, speed and time information; IMU measures the acceleration, angular velocity and attitude of the ship; ultrasonic sensors are used for short-range obstacle detection and ranging; barometers measure atmospheric pressure to assist altitude measurement and weather forecasting; infrared thermal imagers detect other ships and obstacles in low visibility conditions; radar systems detect other ships, land and obstacles at a long distance; underwater sonar systems detect underwater obstacles and seabed topography; AIS receivers receive ship identification and navigation information sent by other ships and shore base stations.

[0091] The signal conditioning module also includes a preamplifier, an isolation circuit, a temperature compensation system and a power supply module. The programmable filter is a bandpass filter with adjustable bandwidth, wherein: the preamplifier integrates an AGC function to adjust the gain according to the input signal strength; the temperature compensation system compensates for the impact of temperature changes on the signal; and the power supply module provides multiple power supplies.

[0092] Specifically, the preamplifier uses AD8367, which has a built-in AGC function, the isolation circuit uses ADuM1401, the temperature compensation system uses ADT7420, the programmable filter uses LTC1068, the gain amplifier uses AD8338, the analog-to-digital converter uses ADS1258, and the power module uses ADP5052.

[0093] In this embodiment, the schematic diagram of the AGC circuit implemented by AD8367 is as follows Figure 2 As shown, the functional block diagram of ADuM1401 is as follows Figure 3 The functional block diagram of ADT7420 is shown in Figure 4 As shown, the filter circuit principle of LTC1068 is as follows Figure 5 As shown, the gain amplifier circuit of AD8338 is as follows Figure 6 As shown, the AD conversion interface circuit of ADS1258 is as follows Figure 7 As shown in Figure 1, the power application circuit of ADP5052 is as follows: Figure 8 shown.

[0094] In this embodiment, the signal processing flow is:

[0095] 1. The signal acquisition device captures various physical quantities. 2. The preamplifier receives the original signal and performs preliminary amplification and AGC processing. 3. The signal passes through the isolation circuit to eliminate common-mode noise. 4. The temperature compensation system adjusts the circuit parameters in real time to offset the temperature effect. 5. The programmable filter performs fine signal filtering. 7. The gain amplifier adjusts the signal to the appropriate amplitude range. 8. Finally, the analog-to-digital converter converts the processed analog signal into a digital signal.

[0096] Specifically, in one embodiment of the present invention, the logic control module integrates a DSP and an embedded microprocessor, and the logic control module has built-in adaptive filtering algorithm, fuzzy control algorithm, PID control algorithm, and temperature compensation algorithm, wherein:

[0097] The adaptive filtering algorithm is configured in the DSP to control the programmable filters and adjust the parameters of each filter;

[0098] The fuzzy control algorithm is configured in the embedded microprocessor for AGC control of the preamplifier;

[0099] The PID control algorithm is configured in the FPGA and used to control the gain amplifier;

[0100] The temperature compensation algorithm is configured in the embedded microprocessor and is used to optimize the temperature compensation model in the temperature compensation system.

[0101] In this embodiment, the logic control module adopts the control chip of ZYNQ7000. ZYNQ7000 integrates a dual-core ARMCortex-A9 processor, programmable logic and hard-core DSP Slices. Reconfigurable data flow pipeline and PID control are implemented in the FPGA part, and adaptive filtering is implemented by DSP Slices. The dual-core ARM Cortex-A9 processor is used as an embedded microprocessor, and fuzzy control and temperature compensation algorithms are run on the embedded microprocessor. Specifically, the programmable logic is FPGA programmable logic, and the reconfigurable data flow pipeline allows the system to dynamically adjust the data processing flow as needed, thereby improving the flexibility and efficiency of the system. Specifically, the internal structure of ZYNQ7000 is as follows: Fig. 9 shown.

[0102] The algorithms are allocated as follows:

[0103] Adaptive filtering algorithm: configured in DSP, used to control programmable filters and dynamically adjust filter parameters.

[0104] Fuzzy control algorithm: runs on the ARM Cortex-A9 processor and is used to control the AGC (automatic gain control) of the preamplifier and the parameter adjustment of the anti-aliasing filter.

[0105] PID control algorithm: implemented in the FPGA part, used to accurately control the gain amplifier.

[0106] Temperature compensation algorithm: runs on the ARM Cortex-A9 processor to optimize the temperature compensation model.

[0107] Specifically, the expression of the temperature compensation model is:

[0108] S c =S m ×(1+α(TT r )+β(TT r ) 2 )+γ

[0109] In the formula, S c is the signal value after temperature compensation, S m is the measured original signal value, T is the current temperature, T r is the reference temperature, α is the first-order temperature coefficient, β is the second-order temperature coefficient, and γ is the offset correction factor.

[0110] The model takes the form of a second-order polynomial that takes into account the linear and nonlinear effects of temperature on the signal, as well as possible offsets.

[0111] like Fig.10 As shown, the temperature compensation algorithm is:

[0112] Step 1: Set the initial parameters α, β, γ; set the learning rate η to control the amplitude of parameter update in each iteration; set the number of iterations N max ;

[0113] Step 2: Collect signal acquisition device data and corresponding true values ​​at different temperatures;

[0114] Step 3: Set the number of iterations = 1;

[0115] Step 4. Select a data point x i ;

[0116] Step 5: Calculate the temperature compensation model for the data point x i The predicted value of

[0117]

[0118] Step 6: According to the predicted value and data point x i The true value of Calculation error E(x i ):

[0119]

[0120] Step 7: Update the parameters α, β, and γ according to the error:

[0121] α=α-η×v α

[0122] β=β-η×v β

[0123] γ=γ-η×v γ

[0124]

[0125] In the formula, v α 、v β 、v γ is the momentum term, μ is the momentum coefficient, represents partial derivative;

[0126] Step 8: Repeat steps 4 to 7 to traverse all data points;

[0127] Step 9: Determine whether the number of iterations reaches N max If yes, then output the trained temperature compensation model; if no, set the number of iterations +1 and return to step 4;

[0128] Step 10: Use the trained temperature compensation model to perform real-time temperature compensation;

[0129] Step 11: Periodically collect new data and use the new data to fine-tune the temperature compensation model.

[0130] This temperature compensation algorithm allows the system to adapt to environmental changes through continuous learning after initial training, improving the accuracy and robustness of temperature compensation.

[0131] Specifically, in one embodiment of the present invention, the hardware facilities of the data analysis layer include a deep learning accelerator, a neural network inference engine, and a processor, wherein the deep learning accelerator may use NVIDIA DLA, the neural network inference engine may be integrated in the deep learning accelerator, and the processor may use an Intel Xeon D-2700 series processor. The execution process of the data analysis layer is as follows:

[0132] Receive processed data from the signal processing layer and perform preliminary data cleaning and formatting to ensure that the data meets the analysis requirements.

[0133] Start the algorithm selection mechanism. The algorithm selection mechanism is as follows:

[0134] Scan the input data and extract key meta-features, including data type, scale, dimension, and sparsity. According to the current navigation status and preset goals, determine the analysis tasks (such as collision risk assessment, route optimization, etc.). According to the preset goals and data characteristics, classify the tasks.

[0135] Based on the task type and key meta-features of the data, a predefined heuristic rule is used to perform preliminary screening to obtain the first set of candidate algorithms;

[0136] Using historical performance data and meta-learning models, combined with current resources, the first candidate algorithm set is screened again to obtain a second candidate algorithm set;

[0137] The comprehensive score of each algorithm in the second candidate algorithm set is calculated according to the scoring formula, and the algorithm with the highest comprehensive score is selected as the optimal algorithm.

[0138] The execution process of the predefined heuristic rules is as follows:

[0139] Design predefined heuristic rules, including rules based on task type and rules based on data features;

[0140] Initialize an empty set of candidate algorithms;

[0141] According to the task type, add the corresponding algorithm to the candidate algorithm set; for example, for classification tasks, select SVM, random forest, neural network, etc.

[0142] Apply the rules one by one based on the data characteristics:

[0143] If the data scale is large, keep the algorithm that can process big data;

[0144] If the data is high-dimensional, add a dimensionality reduction algorithm;

[0145] If the data is highly nonlinear, the nonlinear algorithm is retained;

[0146] If the data categories are unbalanced, add algorithms to handle unbalanced data;

[0147] If the data is sparse, keep the algorithm that handles sparse data;

[0148] Assign preliminary priorities to candidate algorithms based on the degree of rule matching;

[0149] Set the maximum candidate algorithm number threshold N and the matching degree threshold p, sort the candidate algorithms according to the matching degree from large to small, first filter the candidate algorithms according to the matching degree threshold p, if the number of filtered candidate algorithms does not exceed the maximum candidate algorithm number threshold N, then use the filtered candidate algorithms as the first candidate algorithm set; if the number of filtered candidate algorithms exceeds the maximum candidate algorithm number threshold N, then retain the top N algorithms as the first candidate algorithm set according to the preliminary priority.

[0150] In this embodiment, a specific example is used for illustration:

[0151] Quantify data features into specific values: data size: small (1), medium (2), large (3), very large (4); dimension: low (1), medium (2), high (3), very high (4); degree of nonlinearity: linear (1), weak nonlinearity (2), strong nonlinearity (3); degree of category imbalance: balanced (1), slightly imbalanced (2), severely imbalanced (3); sparsity: dense (1), moderately sparse (2), highly sparse (3).

[0152] Design a scoring matrix. Create a scoring matrix for each rule and each algorithm. For example:

[0153] Algorithms / Rules Big Data Processing High Dimensional Data Nonlinear Class Imbalance Sparse Data SVM 2 4 3 2 4 Random Forest 4 3 4 3 3 Neural Networks 3 5 5 3 2 ...

[0154] Among them, 5 means very suitable and 1 means not very suitable.

[0155] According to the current task, a weight is assigned to each rule, which is as follows: 0.3 for big data processing, 0.2 for high-dimensional data, 0.2 for nonlinearity, 0.15 for class imbalance, and 0.15 for sparse data.

[0156] For each algorithm, calculate its matching degree with the current data features: matching degree = Σ(rule weight * scoring matrix score * data feature level).

[0157] For example, for the SVM algorithm, if the current data features are: large scale (3), high dimension (3), weak nonlinearity (2), slightly unbalanced (2), and moderately sparse (2), then the SVM matching degree = 0.3*2*3+0.2*4*3+0.2*3*2+0.15*2*2+0.15*4*2=7.2.

[0158] After calculating the matching degree of each algorithm, normalization can be performed to adjust the matching degree to a range of 0-1 for comparison.

[0159] According to the normalized matching value, the algorithm is assigned a preliminary priority, and then the algorithms are sorted from large to small according to the preliminary priority. The maximum candidate algorithm number threshold N and the matching threshold p are set, for example, N = 10, p = 0.7, and the algorithms with a matching degree greater than or equal to 0.7 are screened out to form an algorithm list. Then the number of algorithms in the algorithm list is counted. If it does not exceed 10, the algorithms in the algorithm list are used as the first candidate algorithm set. If the number of algorithms in the algorithm list exceeds 10, the first 10 algorithms are retained as the first candidate algorithm set.

[0160] Specifically, the process of secondary screening using historical performance data and meta-learning models is as follows:

[0161] First, build a performance database to record the performance indicators (such as accuracy, running time, memory usage, etc.) of each algorithm under different tasks and data characteristics, and store task characteristics, data characteristics, algorithm parameters and corresponding performance results.

[0162] Using task features and data features as retrieval conditions, the similarity between the current task and historical tasks is calculated using methods such as cosine similarity or Euclidean distance, and the Top-K most similar historical tasks are selected.

[0163] For each candidate algorithm, calculate its average performance in similar tasks and generate a performance ranked list.

[0164] The task features, data features and algorithm features are combined into a meta-feature vector, and the features are normalized and standardized.

[0165] The meta-learning model is trained using historical data, with the model input being the meta-feature vector and the output being the algorithm performance prediction.

[0166] Use the trained meta-learning model to predict the performance of each candidate algorithm on the current task and generate a ranked list of predicted performance.

[0167] Monitor system resource status (CPU usage, available memory, disk I / O, etc.) in real time and evaluate currently available computing resources.

[0168] Based on historical data and current task scale, the resource requirements of each candidate algorithm are predicted and a resource requirement score list is generated.

[0169] A weighted scoring function is set to perform a weighted sum of historical performance, predicted performance, and resource requirements, and the weighted scoring function is applied to each candidate algorithm to generate a weighted scoring list.

[0170] Define the minimum acceptable score and the maximum number of candidate algorithms M, sort the candidate algorithms from high to low by score, and select the top M algorithms with scores above the threshold.

[0171] Generate a detailed list including algorithm name, expected performance, and resource requirements, i.e., the second candidate algorithm set.

[0172] In this embodiment, after the algorithms are screened twice, there may still be more than one remaining algorithm. In this case, the final score of the remaining algorithms is calculated according to a scoring formula. Since the remaining algorithms are highly similar in all aspects after two screenings, the scoring formula introduces a difference adjustment function as well as adaptability score, explainability score, stability score, etc. to more carefully evaluate the performance of each algorithm. Specifically, the scoring formula is as follows:

[0173] F=G(P,E,R,A,I,S)+D(M,L)

[0174]

[0175] D(M,L)=ω m ×M+ω l ×L

[0176] Where F is the comprehensive score, G(P,E,R,A,I,S) is the main scoring function, P is the historical performance score, E is the computational efficiency score, R is the resource usage score, A is the adaptability score, I is the interpretability score, S is the stability score, δ,ε,∈,σ,ξ,κ are dynamically adjusted parameters, D(M,L) is the difference adjustment function, M is the complexity control factor, L is the learning adaptability, ω m ,ω l is an adjustable trade-off parameter.

[0177] In this embodiment, Among them, P i is the i-th historical performance score, w i is the corresponding weight; Among them, T e is the algorithm execution time, T b is the benchmark execution time; Among them, CPU used and Mem usedThey are CPU and memory usage, CPU total and Mem total is the total available quantity; Among them, Per test is the test set performance, Per train is the training set performance; Among them, N if is the number of interpretable features, N tf is the total number of features, M t It is the transparency of the model, with a value between 0 and 1; Among them, P1, P2, ..., P n represents the performance score of n runs, SD represents the standard deviation, and Mean represents the mean; Among them, T exec is the algorithm execution time, T max is the maximum allowed time, Mem used is the amount of memory used by the algorithm, Mem max is the maximum allowed memory; Among them, P new is the performance of the algorithm on new data, P old is the performance of the algorithm on old data.

[0178] If a machine learning or deep learning algorithm is selected, the corresponding pre-trained model can be called from the model library, and the deep learning accelerator in the edge processing unit can be used for fast model fine-tuning.

[0179] Comprehensively analyze data from different sensors, for example, combine GPS location data and radar detection data for more accurate collision risk assessment, analyze hull stability based on IMU data, and predict sailing conditions based on meteorological data. Input the prepared data into the selected algorithm or model. For deep learning models, use the neural network inference engine for efficient calculations. For traditional algorithms, execute on the processor of the edge processing unit.

[0180] The real-time decision-making unit receives the analysis results and generates specific decision suggestions based on the preset decision rules. For example, it provides route adjustment suggestions, issues collision warnings, etc.

[0181] The analysis results and decision recommendations are transmitted to the control communication layer. At the same time, key information during the execution process (such as algorithm performance, resource usage, etc.) is recorded for future algorithm selection optimization.

[0182] Regularly evaluate the effectiveness of the algorithm selection mechanism and update the meta-learning model and historical performance database.

[0183] Specifically, in one embodiment of the present invention, the system control unit uses an ARM processor, such as ARM Cortex-A72. The system control unit switches the system working mode according to the host computer instruction, such as normal navigation, emergency obstacle avoidance, energy-saving mode, etc. The unit monitors the system status in real time and reports to the host computer. The system control unit is configured with a resource scheduler and a power consumption manager. The resource scheduler schedules resources to each subsystem or module according to the resource allocation instruction. The power consumption manager monitors the power consumption of each component and adjusts the system power consumption according to the working mode and environmental conditions.

[0184] The data transmission module is responsible for data exchange within the system and communication with external systems. It includes a multi-address transmission controller to support multiple communication protocols, such as Ethernet, CAN, UART, etc. The data compression unit uses a compression algorithm to compress the transmission data in real time. The compression algorithm can be selected as LZ4, and the compression strategy can be selected according to the data type. The encryption transmission module has a built-in encryption algorithm, such as AES-256, to protect sensitive data and achieve secure key management and exchange. The storage management unit is responsible for the storage and management of system data, which includes a cache and a large-capacity storage interface. The cache uses a high-performance SSD as a cache and supports both write-back and direct write modes. The large-capacity storage interface supports large-capacity SSDs and implements RAID to improve data read and write performance. The storage management unit uses hierarchical storage, automatically adjusts the storage location according to the importance and access frequency of the data, and is equipped with a data backup and recovery mechanism to ensure the security of critical data.

[0185] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-channel signal acquisition and analysis system for safe navigation of ships, It is characterized in that include: The signal acquisition layer includes signal acquisition equipment and signal conditioning modules. The signal acquisition equipment includes GPS, IMU, ultrasonic sensor, barometer, infrared thermal imager, radar system, underwater sonar system, AIS receiver, and the signal conditioning module includes programmable filter, gain amplifier and analog-to-digital converter. The signal processing layer includes a logic control module, which uses FPGA and a reconfigurable data flow pipeline to perform synchronous signal processing; The data analysis layer includes an edge processing unit, an analysis algorithm library, and a real-time decision unit. The edge processing unit uses a deep learning accelerator and a neural network inference engine for data analysis. The analysis algorithm library contains various algorithms and is equipped with an algorithm selection mechanism. The real-time decision unit generates real-time decisions based on the data analysis results. The control and communication layer includes a system control unit, a data transmission module and a storage management unit. The system control unit manages the working mode according to the instructions of the upper computer and is configured with a resource scheduler and a power consumption manager. The data transmission module is configured with a multi-address transmission controller, a data compression unit and an encryption transmission module. The storage management unit uses a high-speed cache and a large-capacity storage interface to store data.

2. A multi-channel signal acquisition and analysis system for safe navigation of ships as claimed in claim 1, It is characterized in that The signal conditioning module also includes a preamplifier, an isolation circuit, a temperature compensation system and a power supply module. The programmable filter is a bandpass filter with adjustable bandwidth, where: The preamplifier integrates AGC function to adjust the gain according to the input signal strength; the temperature compensation system compensates for the impact of temperature changes on the signal; the power module provides multi-channel power supply.

3. A multi-channel signal acquisition and analysis system for safe navigation of ships as claimed in claim 2, It is characterized in that The logic control module integrates DSP and embedded microprocessor. The logic control module has built-in adaptive filtering algorithm, fuzzy control algorithm, PID control algorithm and temperature compensation algorithm, among which: The adaptive filtering algorithm is configured in the DSP to control the programmable filters and adjust the parameters of each filter; The fuzzy control algorithm is configured in the embedded microprocessor for AGC control of the preamplifier; The PID control algorithm is configured in the FPGA and used to control the gain amplifier; The temperature compensation algorithm is configured in the embedded microprocessor and is used to optimize the temperature compensation model in the temperature compensation system.

4. A multi-channel signal acquisition and analysis system for safe navigation of ships as claimed in claim 3, It is characterized in that The expression of the temperature compensation model is: S c =S m ×(1+α(TT r )+β(TT r ) 2 )+c In the formula, S c is the signal value after temperature compensation, S m is the measured original signal value, T is the current temperature, T r is the reference temperature, α is the first-order temperature coefficient, β is the second-order temperature coefficient, and γ is the offset correction factor.

5. A multi-channel signal acquisition and analysis system for safe navigation of ships as claimed in claim 4, It is characterized in that The temperature compensation algorithm is: Step 1: Set the initial parameters α, β, γ; set the learning rate η to control the amplitude of parameter update in each iteration; set the number of iterations N max ; Step 2: Collect signal acquisition device data and corresponding true values ​​at different temperatures; Step 3: Set the number of iterations = 1; Step 4. Select a data point x i ; Step 5: Calculate the temperature compensation model for the data point x i The predicted value of Step 6: According to the predicted value and data point x i The true value of Calculation error E(x i ): Step 7: Update the parameters α, β, and γ according to the error: α=α-η×v α β=β-η×v β γ=γ-η×v γ In the formula, v α 、v β 、v γ is the momentum term, μ is the momentum coefficient, represents partial derivative; Step 8: Repeat steps 4 to 7 to traverse all data points; Step 9: Determine whether the number of iterations reaches N max If yes, then output the trained temperature compensation model; if no, set the number of iterations +1 and return to step 4; Step 10: Use the trained temperature compensation model to perform real-time temperature compensation; Step 11: Periodically collect new data and use the new data to fine-tune the temperature compensation model.

6. A multi-channel signal acquisition and analysis system for safe navigation of ships as claimed in claim 1, It is characterized in that The algorithm selection mechanism is as follows: Scan the input data to extract key meta-features, including data type, scale, dimension, and sparsity, and classify the tasks according to the preset goals; Based on the task type and key meta-features of the data, a predefined heuristic rule is used to perform preliminary screening to obtain the first set of candidate algorithms; Using historical performance data and meta-learning models, combined with current resources, the first candidate algorithm set is screened again to obtain a second candidate algorithm set; The comprehensive score of each algorithm in the second candidate algorithm set is calculated according to the scoring formula, and the algorithm with the highest comprehensive score is selected as the optimal algorithm.

7. A multi-channel signal acquisition and analysis system for safe navigation of ships as claimed in claim 6, It is characterized in that The execution process of the predefined heuristic rules is as follows: Design predefined heuristic rules, including rules based on task type and rules based on data features; Initialize an empty set of candidate algorithms; According to the task type, add the corresponding algorithm to the candidate algorithm set; Apply the rules one by one based on the data characteristics: If the data scale is large, keep the algorithm that can process big data; If the data is high-dimensional, add a dimensionality reduction algorithm; If the data is highly nonlinear, the nonlinear algorithm is retained; If the data categories are unbalanced, add algorithms to handle unbalanced data; If the data is sparse, keep the algorithm that handles sparse data; According to the matching degree of the rule matching, the candidate algorithms are given a preliminary priority; Set the maximum candidate algorithm number threshold N and the matching degree threshold p, sort the candidate algorithms according to the matching degree from large to small, first filter the candidate algorithms according to the matching degree threshold p, if the number of filtered candidate algorithms does not exceed the maximum candidate algorithm number threshold N, then use the filtered candidate algorithms as the first candidate algorithm set; if the number of filtered candidate algorithms exceeds the maximum candidate algorithm number threshold N, then retain the top N algorithms as the first candidate algorithm set according to the preliminary priority.

8. A multi-channel signal acquisition and analysis system for safe navigation of ships as claimed in claim 6, It is characterized in that The scoring formula is as follows: F=G(P,E,R,A,I,S)+D(M,L) D(M,L)=ω m ×M+ω l ×L Where F is the comprehensive score, G(P,E,R,A,I,S) is the main scoring function, P is the historical performance score, E is the computational efficiency score, R is the resource usage score, A is the adaptability score, I is the interpretability score, S is the stability score, δ,ε,∈,σ,ξ,κ are dynamically adjusted parameters, D(M,L) is the difference adjustment function, M is the complexity control factor, L is the learning adaptability, ω m ,ω l is an adjustable trade-off parameter.

9. A multi-channel signal acquisition and analysis system for safe navigation of ships as claimed in claim 3, It is characterized in that The preamplifier uses AD8367, which has a built-in AGC function, the isolation circuit uses ADuM1401, the temperature compensation system uses ADT7420, the programmable filter uses LTC1068, the gain amplifier uses AD8338, the analog-to-digital converter uses ADS1258, and the power module uses ADP5052.

10. A multi-channel signal acquisition and analysis system for safe navigation of ships as claimed in claim 9, It is characterized in that The logic control module adopts the control chip of ZYNQ7000, which integrates dual-core ARM Cortex-A9 processor, programmable logic and hard-core DSP Slices. Reconfigurable data flow pipeline and PID control are implemented in the FPGA part, and adaptive filtering is realized by DSPSlices. The dual-core ARM Cortex-A9 processor is used as an embedded microprocessor, and fuzzy control and temperature compensation algorithms are run on the embedded microprocessor.

Images