An adaptive adjustment method and system for the rheological characteristics of intelligent loading machines in ports

Abstract

This invention relates to the field of intelligent port logistics equipment technology, specifically to a method and system for adaptive adjustment of material rheological characteristics in a port intelligent loading machine. The invention includes the following steps: associating the physical properties and rheological characteristics of the material to obtain a basic rheological characteristic index; quantifying the velocity field and shear rate using optical flow methods to obtain dynamic mechanical characteristics; fusing current rheological parameters, the basic rheological characteristic index, and the dynamic mechanical characteristics using a neural network to obtain a comprehensive rheological state index; analyzing external influencing factors based on environmental and operational influences using environmental impact coefficients and operational state indices; and optimizing operational parameters in real time through reinforcement learning based on rheological trends, the comprehensive rheological state index, and the external influencing factors to achieve adaptive adjustment of rheological characteristics. This invention achieves precise control of material rheological characteristics through the integration of multiple technologies, effectively promoting the transformation of port loading from experience-driven to data-driven approaches.

Description

Technical Field

[0001] This invention relates to the field of intelligent port logistics equipment technology, specifically to an adaptive adjustment method and system for the rheological characteristics of a port intelligent loading machine. Background Technology

[0002] In port bulk material loading operations, the rheological properties of materials (such as viscosity, yield stress, and shear rate response) are influenced by multiple factors including physical properties, environmental conditions, and operating parameters, exhibiting significant nonlinear and time-varying characteristics. Traditional port loading machines often use fixed operating parameters or rely on manual experience for adjustment, which presents the following core problems: One-sided data perception: Relying on a single sensor (such as a pressure sensor) to obtain local data cannot fully reflect the dynamic changes in the rheological properties of materials, and is prone to lag in regulation due to data blind spots; Poor model adaptability: Using a fixed mathematical model to describe the rheological laws of materials makes it difficult to cope with the differences in characteristics of different batches of materials (such as wet coal and dry coal, coarse-grained ore and fine-powdered ore) and extreme environments (high temperature and high humidity, strong winds). Insufficient adjustment precision: Lack of coupled analysis of external factors such as environment and operation. When multiple factors interfere together, problems such as material blockage, spillage or low loading efficiency are likely to occur. According to statistics, the material blockage rate of traditional loading machines is as high as 8%-12%, which seriously affects the continuity of operation. Lack of learning ability: The system cannot utilize historical operational data to optimize adjustment strategies, requiring manual parameter recalibration when facing new materials or operating conditions, resulting in extremely poor adaptability. Current technologies only achieve basic adaptation through the correlation adjustment of speed and flow rate, without addressing deep perception and dynamic optimization of material rheological characteristics. While visual data is introduced, it is not combined with multi-source sensor fusion and online learning, leaving significant shortcomings in adjustment accuracy and robustness. Therefore, there is an urgent need for an adaptive adjustment algorithm for material rheological characteristics that can comprehensively perceive, accurately model, and dynamically learn, to address the intelligent bottleneck in port loading operations. Summary of the Invention

[0003] To address the shortcomings of existing methods and the needs of practical applications, and to solve the aforementioned problems, this invention provides an adaptive adjustment method for the rheological properties of materials in a port intelligent loading machine. The method includes the following steps: associating the physical properties and rheological characteristics of the material to obtain a basic rheological property index; quantifying the velocity field and shear rate using optical flow methods to obtain dynamic mechanical characteristics; fusing current rheological parameters, the basic rheological property index, and the dynamic mechanical characteristics using a neural network to obtain a comprehensive rheological state index; analyzing external influencing factors based on environmental and operational influences using an environmental impact coefficient and an operational state index; and optimizing operational parameters in real time through reinforcement learning based on rheological trends, the comprehensive rheological state index, and the external influencing factors to achieve adaptive adjustment of rheological properties.

[0004] This invention correlates the physical properties and rheological characteristics of materials to obtain a basic index, anchoring the inherent rheological properties of materials and providing a benchmark for subsequent adjustments, thus solving the problem of blind adjustments caused by the differences in characteristics between different batches of materials. It utilizes optical flow methods to quantify the velocity field and shear rate to obtain dynamic characteristics, capturing the material's flow dynamics in real time, compensating for the shortcomings of static properties in reflecting changes in operating conditions, and providing data support for accurate judgment of rheological dynamics. Through neural networks, it fuses multiple features to obtain a comprehensive index, integrating static basic, dynamic mechanical, and real-time parameters to generate a unified rheological state assessment index, realizing a shift from fragmented perception to holistic judgment, and accurately identifying safe / warning / dangerous states. Based on the analysis of external influencing factors of the environment and operation, it quantifies the comprehensive effect of external disturbances on rheology, avoiding adjustment deviations caused by isolated evaluation of single factors, making the adjustment strategy more aligned with actual operating conditions. Through reinforcement learning, it optimizes operating parameters, combining rheological trends and comprehensive evaluation results for dynamic adjustment, achieving predictive optimization, solving the problem of traditional adjustment lag, reducing material blockage rate, and improving loading efficiency and operational safety.

[0005] Optionally, the physical properties and rheological characteristics of the associated materials are used to obtain a basic rheological property index, which includes the following steps: This invention acquires sensor data and preprocesses it. From the preprocessed stationary data, key features are extracted using a combination of statistical analysis and feature engineering. A dynamic weighted fusion algorithm effectively integrates multiple features, and the fused comprehensive feature vector is used to calculate the basic rheological property index. This invention cleanses sensor noise and interference through data preprocessing, synchronizes multi-source data, and unifies the dimensions, providing a reliable data base for subsequent analysis. The combination of statistics and feature engineering extracts key features, eliminates redundant information, and accurately captures the core dimensions of material physical properties. Dynamic weighted fusion adapts to differences in material characteristics, achieving efficient integration of multiple features. The output basic rheological index quantifies the inherent flow characteristics of the material, providing an attribute benchmark for subsequent adjustments, solving the problem of blind adjustments caused by differences in batch characteristics of different materials, and ensuring that subsequent dynamic adjustments are more targeted.

[0006] Optionally, the step of quantizing the velocity field and shear rate using the optical flow method to obtain dynamic mechanical characteristics includes the following steps: Based on material flow data, this invention calculates the optical flow field and shear rate distribution; it extracts core features from high-dimensional velocity field and shear rate data to obtain dynamic mechanical characteristics. This invention accurately quantifies the velocity field and shear rate distribution of material flow using optical flow methods, reflecting real-time changes in the material flow dynamics state and avoiding the adjustment lag caused by traditional static sensing. It extracts key features from high-dimensional flow field data, transforming complex flow information into quantifiable dynamic mechanical characteristics, providing dynamic input for subsequent neural network fusion analysis, and achieving full-dimensional perception of both static foundation and dynamic process. The dynamic mechanical characteristics are directly related to the real-time fluctuations of material rheological properties, providing a dynamic basis for the synthesis of comprehensive rheological state indices, helping to accurately identify flow risks and improve the targeting and effectiveness of adaptive regulation.

[0007] Optionally, obtaining the comprehensive rheological state index by fusing the current rheological parameters, the basic rheological property index, and the dynamic mechanical characteristics using a neural network includes the following steps: The present invention standardizes the current rheological parameters, the basic rheological property index, and the dynamic mechanical characteristics; constructs a fully connected neural network structure; and obtains a comprehensive rheological state index based on the standardization results through the fully connected neural network structure. Standardization eliminates dimensional differences and numerical range interference among current rheological parameters, basic indices, and dynamic mechanical characteristics, ensuring balanced weighting of features from different sources during fusion. This prevents a single feature from dominating the results due to numerical advantages, laying a data foundation for accurate fusion. Fully connected neural networks can capture nonlinear correlations between multiple features, integrating static basic attributes, dynamic flow characteristics, and real-time rheological parameters into a unified comprehensive index. This achieves a shift from fragmented perception to holistic evaluation, accurately quantifying material flow stability and providing clear and reliable state basis for subsequent reinforcement learning adjustments.

[0008] Optionally, obtaining the current rheological parameters includes the following steps: Calculate shear stress; select a constitutive model based on a triple selection mechanism of basic index grading, material type adaptation, and dynamic adjustment under operating conditions; combine the shear stress and the constitutive model to solve for the current rheological parameters. Calculating shear stress provides accurate mechanical input for solving rheological parameters; data purification and correction eliminate interference, ensuring reliable basic data; the triple mechanism for model selection solves the problem of poor adaptability of traditional fixed models, accurately matching different materials (low / medium / high viscosity) and operating conditions, improving the relevance of parameter calculations; solving for rheological parameters outputs quantified real-time rheological characteristics of the material, providing a core basis for subsequent comprehensive state assessment and reinforcement learning adjustments, ensuring that adjustment strategies align with material dynamics, reducing clogging risks, and improving operational stability.

[0009] Optionally, the analysis of external influencing factors based on environmental and operational impacts using environmental impact coefficients and operational status indices includes the following steps: A collaborative impact assessment model is constructed. Based on the environmental impact coefficient and the operational state index, initial external impact factors are calculated using the collaborative impact assessment model. These initial external impact factors are then corrected using rheological trend prediction results to obtain external impact factors based on both environmental and operational influences. This invention calculates initial factors through a collaborative model, quantifies environmental-operational coupled interference, overcomes the bias of traditional isolated assessments of single factors, and avoids the neglect of the risks associated with the superposition of these two factors, providing a scientific quantitative framework for external impact analysis. Combined with dynamic rheological trend optimization assessment, factors are aligned with the real-time state of materials, avoiding the disconnect between static data and actual operating conditions, improving accuracy, providing a core reference for reinforcement learning regulation, ensuring that regulation strategies adapt to comprehensive interference, enhancing the targeting of regulation, and helping to reduce congestion risks and improve operational stability.

[0010] Optionally, the environmental impact coefficient is extracted by the following steps: The environmental variables are fuzzified and a rule base is constructed; fuzzy inference is performed based on the rule base and the fuzzy results, and the inference results are defuzzified to obtain the environmental impact coefficient; Extracting the operational status index includes the following steps: Analyze the operating parameters and evaluate the disturbance intensity; use the operating parameters and the disturbance intensity to synthesize the operating state index.

[0011] This invention addresses the challenge of accurately assessing environmental impacts by fuzzifying environmental variables, constructing a rule base, and employing fuzzy inference. This transforms nonlinear environmental factors such as temperature and humidity into quantifiable coefficients, providing a quantitative basis for subsequent regulation. Furthermore, by analyzing operating parameters and evaluating disturbance intensity, combined with a dynamic weighted index synthesis, the invention precisely quantifies the impact of operating parameters deviating from optimal values, avoiding the limitations of single-parameter assessments. Together, these two methods achieve comprehensive quantification of external influences, providing reliable input for integrated rheological assessment and reinforcement learning regulation, thereby enhancing the targeting and accuracy of regulation strategies.

[0012] Optionally, the step of optimizing operating parameters in real time through reinforcement learning based on rheological trends, the comprehensive rheological state index, and the external influencing factors to achieve adaptive adjustment of rheological properties includes the following steps: Based on the rheological trend, the comprehensive rheological state index, and the external influencing factors, a state space is defined, and an action space and reward function are designed. Based on the state space, action space, and reward function, operational parameters are optimized in real time through reinforcement learning to achieve adaptive adjustment of rheological characteristics. This invention defines the state space using rheological trends, a comprehensive index, and external factors, designs the action space in conjunction with the loading machine's parameter characteristics, and designs the reward function with the goal of reducing risk and stabilizing operating conditions, providing precise input and clear guidance for reinforcement learning. Reinforcement learning optimizes parameters in real time based on multi-source dynamic data, achieving a prediction-adjustment-feedback closed loop and proactively avoiding congestion risks.

[0013] Optionally, analyzing the rheological trend includes the following steps: This invention preprocesses historical rheological data from ports; it constructs an online learning real-time update model based on an online learning algorithm, using this model to predict future rheological characteristic trends from historical rheological data. By preprocessing historical rheological data, outliers are removed and time series are regularized, providing a high-quality data foundation for the model and solving the problem of port data being interfered with by vibration and sensor drift. The online learning model can be updated in real time, quickly adapting to dynamic changes in material batches and the environment, overcoming the adaptation limitations of traditional fixed models. It accurately outputs future rheological trends, enabling early prediction and proactive adjustment, avoiding the lag of traditional adjustments, providing forward-looking basis for optimizing loading machine parameters, helping to reduce congestion risks, and ensuring continuous operation.

[0014] Secondly, to efficiently execute the adaptive adjustment method for material rheological characteristics of a port intelligent loading machine provided by this invention, this invention also provides an adaptive adjustment system for material rheological characteristics of a port intelligent loading machine, including a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. The memory stores a computer program containing program instructions. The processor is configured to call the program instructions to execute the adaptive adjustment method for material rheological characteristics of a port intelligent loading machine as described in the first aspect of this invention. The adaptive adjustment system for material rheological characteristics of a port intelligent loading machine of this invention has a compact structure and stable performance, and can stably execute the adaptive adjustment method for material rheological characteristics of a port intelligent loading machine provided by this invention, further improving the overall applicability and practical application capability of this invention. Attached Figure Description

[0015] Figure 1 A flowchart illustrating an adaptive adjustment method for the rheological characteristics of a port intelligent loading machine, provided in an embodiment of the present invention; Figure 2 This is a framework diagram of an adaptive adjustment system for the material rheological characteristics of a port intelligent loading machine, provided as an embodiment of the present invention. Detailed Implementation

[0016] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0017] Throughout this specification, references to an embodiment, example, or illustration mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, phrases appearing in various places throughout the specification, such as "in one embodiment," "in an embodiment," "an example," or "an illustration," do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in any suitable combination and / or sub-combination in one or more embodiments or examples. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0018] Please see Figure 1 To address the aforementioned problems, this invention provides an adaptive adjustment method for the material rheological characteristics of a port intelligent loading machine, such as... Figure 1 As shown, in one embodiment, the method includes the following steps: S1. Based on the physical properties and rheological characteristics of the associated materials, obtain the basic rheological property index.

[0019] By leveraging the complementarity of data from multiple sensors, errors from a single sensor can be eliminated, allowing for the accurate extraction of the inherent physical properties of materials and laying the foundation for rheological analysis.

[0020] Specifically, the physical properties and rheological characteristics of the associated materials are used to obtain the basic rheological property index, which includes the following steps: S11. Acquire sensor data and preprocess the data.

[0021] In the port operating environment, dust, vibration, and electromagnetic interference can easily lead to distortion of the raw sensor data. Therefore, preprocessing is the first step for reliable data utilization. The denoising process adopts wavelet threshold denoising based on db4 wavelet basis, which effectively filters out high-frequency interference while retaining the data mutation characteristics. Time synchronization is calibrated based on the loading machine PLC master clock to give each sensor data a precise timestamp. Time stamp matching is used to achieve time alignment of multi-source data and avoid fusion deviation caused by data timing misalignment. Range normalization adopts the maximum-minimum normalization method to uniformly map data of different dimensions such as density, moisture content, and particle size distribution to the [0,1] interval, ensuring that the weights of each feature are balanced in subsequent fusion.

[0022] S12. Extract key features from the preprocessed stationary data using a combination of statistical analysis and feature engineering.

[0023] For density data, the mean and variance within a sliding window (100 sampling points) are calculated. The density mean reflects the overall compactness of the material, while the variance reflects local density fluctuations. For moisture content data, the standard deviation and coefficient of variation are extracted. A larger standard deviation indicates a more uneven moisture distribution and a greater likelihood of local agglomeration. For particle size distribution data, the particle size distribution curve output by a laser particle size analyzer is used to extract D10 (cumulative 10% particle size), D50 (cumulative 50% particle size, i.e., median diameter), D90 (cumulative 90% particle size), and the distribution width coefficient ((D90-D10) / D50). D50 directly relates to material flowability; a larger D50 usually indicates poorer material flowability. For hardness data, the coefficient of variation and peak value are calculated. The coefficient of variation reflects the uniformity of material hardness, while the peak value corresponds to the presence of hard impurities in the material. All extracted features are subjected to correlation analysis (Pearson correlation coefficient |r|<0.7) to remove redundant features, forming a key feature set.

[0024] S13. Achieve effective integration of multiple features through a dynamic weighted fusion algorithm, and calculate the basic rheological property index using the fused comprehensive feature vector.

[0025] Weights are determined through historical data error inversion and optimization, constructing an error feedback model. Using manually measured material rheological properties as a benchmark, the mean square error (MSE) between the fused features and the benchmark value is calculated for different weight combinations. The weight combination with the smallest MSE is used as the initial weights. During real-time fusion, weights are dynamically adjusted according to the current state of the material. For example, when the microwave humidity sensor detects a material moisture content below 10% (dry state), the influence of density on rheological properties is significantly enhanced, and the weight of the density feature is increased by 20%. When the moisture content is above 25% (high humidity state), the weight of the moisture content feature is increased. During fusion, each feature is multiplied by its corresponding weight and then summed to form a comprehensive feature vector. This vector fully preserves the core information and interrelationships of each physical property.

[0026] Furthermore, based on the fused comprehensive feature vector, the mapping from physical attributes to rheological properties is completed through a pre-defined physical attribute-rheological correlation model. This correlation model is constructed using multivariate nonlinear regression combined with expert experience, with the comprehensive feature vector as the input and the basic rheological property index as the output.

[0027] During model construction, physical properties and corresponding rheological parameters (viscosity, yield stress) data of various materials (coal, ore, grain, etc.) were collected. After normalization, the data were input into the model for training. The model parameters were optimized using the gradient descent method to ensure that the mean absolute error (MAE) between the model predictions and the measured values ​​was less than 5%. The model clearly defines the correlation between various physical properties and rheological characteristics. For example, particle size D50 is positively correlated with initial viscosity (correlation coefficient r=0.82), and mean density is also positively correlated with initial viscosity (r=0.75). The final output basic rheological property index range is set to 0-100. The higher the index value, the greater the internal friction and the worse the flowability of the material. Among them, 0-30 represents low-viscosity, easily flowable materials, 30-70 represents medium-viscosity materials, and 70-100 represents high-viscosity, difficult-to-flow materials.

[0028] S2. Dynamic mechanical characteristics are obtained by quantifying the velocity field and shear rate using optical flow methods.

[0029] By visually capturing the material flow pattern and combining it with optical flow methods to quantify the velocity field and shear rate, the spatial distribution of the material flow state can be reflected in real time.

[0030] Specifically, the process of quantifying the velocity field and shear rate using optical flow methods to obtain dynamic mechanical characteristics includes the following steps: S21. Calculate the optical flow field and shear rate distribution based on material flow data.

[0031] In port operations, sudden changes in lighting (such as alternation between strong midday light and weak light during rain) can lead to uneven brightness in images. Isolated noise generated by material splashing can easily interfere with subsequent feature extraction. Therefore, preprocessing such as enhancement-denoising-correction is also required.

[0032] First, contrast is enhanced by adaptive histogram equalization (CLAHE) with limited contrast. An 8×8 pixel block size is set for the hopper outlet area (smaller than the conventional 16×16 block, to adapt to the local details of material flow). The contrast limit threshold is set to 2.0 (the threshold for normal scenes is 4.0 to avoid over-enhancing of highly reflective materials in ports). This increases the grayscale difference between the material and the background, significantly distinguishing flowing materials from stationary equipment. Then, a morphological opening operation (erosion followed by dilation) is performed to remove small-area noise. A 3×3 elliptical structural element is selected (adapting to the irregular shape of material splashing, which is better than the excessive erosion of the edges by the square structure). This can effectively filter out dust noise and splash particle residue with an area of ​​less than 5 pixels, while preserving the edge contour of the material flow. Finally, image distortion correction is introduced. Based on the intrinsic parameter matrix of the industrial camera (obtained in advance through checkerboard calibration), the material edge offset caused by lens distortion is corrected to ensure that the image coordinate and physical coordinate error of the connection area between the hopper outlet and the conveyor belt is ≤1mm, laying the foundation for the accuracy of subsequent speed calculation.

[0033] Further, the optical flow field is calculated. First, a Gaussian pyramid (3-5 layers, dynamically adapted to the material flow rate: increasing to 5 layers when the flow rate is >0.5m / s, and reducing to 3 layers when the flow rate is <0.2m / s) is constructed. Coarse-grained optical flow is calculated starting from the top low-resolution image, and iteratively optimized downwards to the original resolution. Second, the core parameters are adjusted: the window size is set to 15×15 pixels. After the calculation is completed, outliers are removed from the obtained velocity vectors (u,v). Error vectors caused by material occlusion (such as vectors with a velocity difference between adjacent pixels >2m / s) are filtered out using the 3σ criterion. Finally, a smooth two-dimensional velocity field is output and visualized using a pseudo-color image (red represents high-speed areas, and blue represents low-speed areas), intuitively presenting the distribution characteristics of high flow velocity at the center and low flow velocity at the edge of the hopper outlet.

[0034] Shear rate is a core indicator reflecting the internal friction intensity of materials and requires accurate calculation based on the velocity field gradient, while also adapting to the non-Newtonian fluid characteristics of port materials. During the calculation, the Sobel operator is used to calculate the partial derivatives of the velocity field (Sobel_x operator in the x-direction and Sobel_y operator in the y-direction), with an operator size of 3×3. Convolution operations are used to convert the velocity vector into a gradient matrix, avoiding noise amplification caused by direct differencing. Subsequently, a 5×5 pixel Gaussian weighted sliding window smoothing is performed (window weights decrease from the center to the edge, with a total weight sum of 1), filtering out high-frequency noise in the gradient calculation while preserving the spatial distribution characteristics of the shear rate, such as the high shear rate clusters appearing in the 10cm area below the hopper outlet due to material compression. Finally, the calculation results are mapped to a shear rate heatmap, and peak areas are marked on the heatmap to provide a basis for subsequent assessment of the risk of material agglomeration.

[0035] S22. Extract core features from high-dimensional velocity field and shear rate data to obtain dynamic mechanical features.

[0036] First, feature types are categorized by function: Velocity field features include regional average velocity (the average magnitude of the velocity vectors of all pixels within the conveyor belt area, reflecting overall flow efficiency), velocity standard deviation (the standard deviation of the magnitude, reflecting flow stability), and velocity gradient peak (the maximum value of the gradient in the x / y direction, reflecting local velocity abrupt changes); shear rate features include shear rate peak (the maximum value in the heatmap, reflecting the extreme value of internal friction in the material), and the proportion of high shear rate regions (shear rate > 200s). -1 The pixel count ratio reflects the range of high-friction areas, and the mean shear rate (overall mean, reflecting average friction intensity) is used. After feature extraction, significant features (confidence > 95%) are screened through analysis of variance, and features with weak correlation to rheological parameters (such as minor features in the velocity field edge region) are removed. The core indicators that are finally retained are standardized to the [0,1] interval (using max-min normalization, and extreme values ​​are obtained from the historical data of 3 months), forming dynamic mechanical features.

[0037] S3. Based on the neural network, the current rheological parameters, the basic rheological property index, and the dynamic mechanical characteristics are fused to obtain the comprehensive rheological state index.

[0038] By combining real-time mechanical data and a suitable constitutive model, the rheological parameters of the current material, such as viscosity and yield stress, are quantified.

[0039] Specifically, obtaining the current rheological parameters includes the following steps: S311. Calculate shear stress.

[0040] First, pressure data preprocessing is performed: The pressure sensor is installed on the inner wall of the loading machine hopper and at the conveyor belt roller. The raw data is affected by mechanical vibration and contains high-frequency noise. A 5-point sliding window weighted filter is used (window weights are [0.1, 0.2, 0.4, 0.2, 0.1], highlighting the weight of intermediate values ​​to suppress instantaneous pulses) to reduce the data fluctuation amplitude. Then, the contact area A is calculated in real time: The preprocessed visual image in step S21 is called, and the contact area between the material and the sensor is segmented using the Otsu threshold segmentation algorithm (the threshold is dynamically adjusted to 1.2 times the grayscale mean to adapt to material color changes). Then, the area is calculated by combining the calibration parameters of the industrial camera (pixel-physical size conversion ratio is 0.1mm / pixel). For example, if the segmented contact area is 8000 pixels, the corresponding actual area A = 8000 × (0.1 × 10⁻⁶) / ( ... -3 m) 2 =8×10 -4 m 2 Finally, force-stress conversion and correction are performed: the shear force F is calculated based on the relationship between force and pressure F=p×S (p is the filtered pressure value, S is the effective sensing area of ​​the sensor), and then the shear stress τ is obtained through τ=F / A.

[0041] S312. Select the constitutive model based on the triple selection mechanism of basic index grading, material type adaptation, and dynamic working condition correction.

[0042] A triple selection mechanism is constructed, consisting of basic index grading, material type adaptation, and dynamic adjustment based on operating conditions, to ensure accurate matching between the constitutive model and the rheological properties of the materials. The core grading logic divides materials into three categories based on basic rheological property indices (0-100): The first category consists of low-viscosity, easily flowable materials with indices < 30 (such as iron ore and dry coal). These materials have no obvious yield stress, and their flow characteristics conform to the power-law model τ=Kγ. n (K is the consistency coefficient, reflecting the viscosity of the material; γ is the shear rate; n is the flow index; n < 1 indicates a pseudoplastic fluid; n = 1 indicates a Newtonian fluid; n > 1 indicates a dilatant fluid.) This model can accurately quantify the impact of flow velocity changes on shear stress. The second category is high-viscosity, difficult-to-flow materials with an index > 70 (such as wet coal slime and bulk fertilizer). These materials need to overcome the initial yield stress to flow, and the Bingham model τ = τ0 + μγ (τ0 is the yield stress, μ is the plastic viscosity) is suitable, which can effectively capture the transition characteristics from static agglomeration to dynamic flow. The third category is medium-viscosity materials with an index of 30-70 (such as semi-wet coal and grain), which have the characteristics of yield stress and viscosity changing with shear rate. The Herschel-Bulkley model τ = τ0 + Kγ is selected. n As the main model, this model is a general form of the first two types of models and has stronger adaptability.

[0043] Simultaneously, a secondary verification of material type is introduced: if the laser particle size analyzer detects D90 > 500 μm, even if the basic index < 30, the lower limit of the flow index n of the power law model will be set to 0.8; if the moisture content > 25%, a humidity correction term will be added to the corresponding model (e.g., the Bingham model is corrected to: τ = τ0 × (1 + 0.02 × W) + μγ, where W is the moisture content), to ensure that the constitutive model is fully adapted to the material characteristics under various working conditions in the port.

[0044] S313. Solve for the current rheological parameters by combining the shear stress and the constitutive model.

[0045] A least squares optimization scheme involving data filtering, weighted fitting, and real-time iteration is adopted to address the fitting accuracy problem caused by fluctuations in τ and γ data during port material flow. First, data filtering is performed: from shear rate distribution and shear stress data, data are filtered according to shear rate intervals [0,50), [50,150), [150,300), and [300,500]. -1 Twenty valid data pairs were extracted evenly (each data pair represents the mean of τ and γ within the interval), and regions with abrupt changes in shear rate (such as the impact zone at the hopper outlet, where γ fluctuations > 100s) were removed. -1 Data ( / 100ms) is used to ensure that the data can reflect the stable flow state of the material.

[0046] Then, weighted least squares fitting is performed: An objective function is designed for different models; for example, for a power-law model, the formula needs to be transformed into a linear form lnτ=lnK+nlnγ by taking the logarithm. The objective function is set as follows: The weight Dynamically allocated based on the data range, shear rate 150-300s -1 (Main flow velocity range) The remaining intervals Highlighting the impact of core flow interval data; the objective function of the Bingham model is For regions with large τ values ​​(τ>τ0+2μγ), the weight is increased to 1.3 to ensure the accuracy of yield stress calculation. The fitting process uses the gradient descent method for iterative solution: initial parameter values ​​are initialized (e.g., K=10, n=0.8 for the power law model; τ0=5Pa, μ=0.5Pa·s for the Bingham model), the learning rate is set to 0.01, and the number of iterations is 500, then the final parameters are output. To meet real-time requirements, the fitting calculation is decomposed into an incremental calculation + cache reuse mode, updating only the fitting results of the latest 10 sets of data each time, controlling the calculation time to within 20ms, and adapting to a 50Hz sampling frequency.

[0047] Furthermore, to ensure the reliability of rheological parameters and avoid adjustment errors caused by model mismatch, residual calculation and evaluation are performed: the mean absolute percentage error (MAPE) and root mean square error (RMSE) are used as dual indicators to quantify the fitting accuracy, and a verification threshold is set: when MAPE < 5% and RMSE < 0.8 Pa, the parameters are considered valid; if 5% ≤ MAPE < 10% or 0.8 Pa ≤ RMSE < 1.5 Pa, parameter correction is triggered; if MAPE ≥ 10% or RMSE ≥ 1.5 Pa, the model is considered to have failed, and the model switching process is initiated. The model switching logic is as follows: when the main model fails, it is preferentially switched to the Herschel-Bulkley general model (applicable to various non-Newtonian fluids), and the parameter solution is re-executed. In addition, after every 10 sets of parameter calculations, cross-validation is performed: the current parameters are used to predict the first 5 sets of historical τ data. If the prediction error increases by 20% compared to the historical average, the data selection interval is adjusted retrospectively (e.g., the data sampling amount in the main velocity interval is expanded) to ensure the long-term stability of the parameters.

[0048] Furthermore, by leveraging the nonlinear fitting capabilities of neural networks and integrating multi-source rheological features, a comprehensive index reflecting the overall rheological state of the material is output.

[0049] Specifically, obtaining the comprehensive rheological state index by fusing the current rheological parameters, the basic rheological property index, and the dynamic mechanical characteristics using a neural network includes the following steps: S321. Standardize the current rheological parameters, the basic rheological property index, and the dynamic mechanical characteristics.

[0050] A standardization strategy combining feature adaptation and outlier anchoring is adopted to uniformly map multi-source features to the [-1,1] interval, which eliminates dimensional differences while preserving the dynamic fluctuation characteristics of port material features. The [-1,1] interval is chosen instead of the traditional [0,1] interval because this interval can more accurately reflect the positive / negative impact of features (e.g., an excessively high shear rate peak indicates positive interference, while an excessively low base index indicates a negative stabilization signal).

[0051] Specifically, the basic rheological property index (0-100) is standardized using a linear offset, with the formula x'=(x-50) / 50. When x=100 (high viscosity), x'=1, and when x=0 (low viscosity), x'=-1, intuitively reflecting the positive and negative effects of viscosity on flow stability; the peak shear rate (0-500s) -1 The offset after extreme value normalization is calculated using the formula x'=((x-x_min) / (x_max-x_min))×2-1, where x_min=20s. -1 x_max=500s -1 (Taken from the statistical extreme values ​​of port coal and iron ore over 3 months), if the peak shear rate of a certain batch of material reaches 600s...-1 (Outliers) are forcibly anchored at x'=1 to prevent extreme data from undermining the standardization effect; rheological parameters are treated differently according to type, and the consistency coefficient K of the power-law model is (1-50 Pa·s). n Log-normalized x' = (lnx - ln25) / (ln50 - ln1) × 2⁻¹, and the flow index n (0.3-1.5) linearly normalized x' = (x - 0.9) / (1.5 - 0.3) × 2⁻¹, are used. The yield stress τ0 (1-30 Pa) of the Bingham model is mapped using (x - 15.5) / 14.5. After all feature processing, a range check is required (99% of the data should fall within [-1, 1]). Otherwise, a sensor data re-check is triggered (e.g., abnormal shear rate may be due to dust accumulation on the camera lens) to ensure the reliability of the input features.

[0052] S322. Construct a fully connected neural network structure, and obtain the comprehensive rheological state index based on the standardized results through the fully connected neural network structure.

[0053] A three-layer fully connected neural network structure consisting of feature enhancement, nonlinear mapping, and output constraints is constructed. Network parameters are optimized to address the coupling characteristics of multi-source features in ports, ensuring sensitivity in identifying stable-abrupt operating conditions. Specifically, the input layer neurons correspond to the standardized feature dimensions: basic rheological property index, regional average velocity, velocity standard deviation, peak shear rate, proportion of high shear rate regions, consistency coefficient K (or yield stress τ0), flow index n (or plastic viscosity μ), and peak velocity gradient. Each neuron is augmented with a batch normalization layer (BN layer) to accelerate training convergence and suppress distribution shifts caused by fluctuations in port data. The hidden layer consists of 16 neurons using the ReLU activation function, which effectively addresses the vanishing gradient problem. Port material characteristics often exhibit local abrupt changes (such as a sudden increase in shear rate), and ReLU can quickly respond to such feature changes. To enhance feature interaction, residual connections are introduced in the hidden layer (directly superimposing 1 / 4 of the input features onto the hidden layer output), avoiding information loss during multi-feature fusion. The output layer uses a sigmoid activation function for one neuron, which strictly constrains the output to the 0-1 range. The comprehensive rheological state index is defined as follows: 0-0.3 is a stable state (uniform material flow, no risk of blockage), 0.3-0.7 is a warning state (requires fine-tuning of parameters), and 0.7-1.0 is a dangerous state (prone to blockage, requires emergency adjustment). The larger the index value, the more unstable the flow, which is precisely matched with the operation response logic of the port loading machine.

[0054] Furthermore, a dataset comprehensively covering port operating conditions is constructed. The model is optimized through a combination of offline pre-training and online fine-tuning to ensure adaptability to various material types and complex operating conditions. The dataset construction includes three core case types: stable loading cases (comprehensive index < 0.3, accounting for 40%, such as uniform flow of dry iron ore), early warning cases (0.3 ≤ comprehensive index < 0.7, accounting for 40%, such as fluctuating flow of semi-wet coal), and dangerous cases (comprehensive index ≥ 0.7, accounting for 20%, such as the state before blockage of wet coal slime). Each case type is further subdivided by material type (coal, iron ore, grain) to avoid model bias towards a single material. During the training phase, the Adam optimizer is used with an initial learning rate of 0.001. A learning rate decay mechanism is introduced, halving the learning rate every 5 epochs (one full iteration of samples). Decay stops when the validation set loss does not decrease for 3 consecutive epochs, balancing training speed and accuracy. The weighted mean squared error (WMSE) loss function is used, with dangerous case samples assigned double the weight. To prevent overfitting, a dropout layer is added between the hidden and output layers, with the dropout probability dynamically adjusted to 0.2 (during training) and 0 (during inference). L2 regularization (weight decay coefficient 0.0001) is also used to constrain the network parameters. After training, the model needs to be validated through real-world testing at a port. A one-month trial run is conducted on three different types of loading machines. Cases with prediction errors > 0.1 (such as predictions of warnings despite actual stability) are included in the incremental dataset, and online fine-tuning is performed (each fine-tuning only updates the parameters of the last three neurons in the output and hidden layers), improving the model's adaptation accuracy to specific loading machines to over 95%.

[0055] In this embodiment, a real-time fusion mechanism of data synchronization, fast inference, and fault tolerance can be constructed to ensure that the exponent is output without delay at a sampling frequency of 50Hz, precisely matching the adjustment response cycle (20ms) of the loading machine. The data synchronization link is triggered by the PLC master clock. When the latest rheological parameters are output (once every 20ms), the basic exponent and visual features are retrieved synchronously to form a real-time feature vector. After standardization, the vector is stored in a 10-frame data buffer pool. If a feature is temporarily missing (e.g., the shear rate data is lost due to camera disconnection), the average feature value of the previous 3 frames in the buffer pool is automatically called to fill the missing feature, avoiding inference interruption. The model inference adopts lightweight optimization, converting the trained neural network into ONNX format and accelerating inference through TensorRT. The single-frame feature inference time is compressed from the original 8ms to 1.2ms, which is far lower than the 20ms cycle requirement. Gradient monitoring is added during inference. If the exponent change rate is >0.15 for 5 consecutive frames, the model confidence check is automatically started. The output result is corrected by combining the shear rate heatmap (if the proportion of high shear rate area is >30%, a dangerous state is confirmed). The final output comprehensive rheological state index is transmitted to the loading machine control system in real time via industrial Ethernet. At the same time, it is visualized on the monitoring interface using three-color lights (green ≤0.3, yellow 0.3-0.7, red ≥0.7), providing operators with an intuitive reference and achieving dual protection of automatic adjustment and manual monitoring.

[0056] S4. Analyze the external influencing factors based on environmental and operational impacts using the environmental impact coefficient and operational status index.

[0057] Fuzzy logic is used to process the nonlinear and uncertain effects of environmental factors (temperature, humidity, etc.) and to map environmental data into quantifiable rheological influence coefficients.

[0058] Specifically, the extraction of the environmental impact coefficient includes the following steps: S411. Fuzzyize environment variables and build a rule base.

[0059] To address the characteristics of continuously changing but ambiguous influence boundaries in port environmental parameters, the quantization ranges and fuzzy set divisions of the three core input variables are first clearly defined to ensure coverage of typical coastal port climate ranges (temperature -5℃~45℃, relative humidity 30%~95%, wind speed 0~10m / s). Temperature is divided into three fuzzy sets: low (-5℃~5℃), medium (5℃~35℃), and high (35℃~45℃); humidity is divided into three levels: dry (30%~50%), humid (50%~75%), and damp (75%~95%); and wind speed is divided into three levels: weak (0~3m / s), medium (3~7m / s), and strong (7~10m / s). A triangular membership function is used to convert precise values ​​to fuzzy values. The general expression of the function is μ(x)=max{0,1-|xc| / d}, where c is the fuzzy set center value and d is the half-width. Taking temperature as an example, the fuzzy set has c=20℃ and d=15℃, meaning the membership degree is 1 in the range of 15℃ to 25℃. The membership degree drops to 0 when the temperature is below 5℃ (20-15) or above 35℃ (20+15), perfectly matching the stable fluidity of port materials (such as coal) around 20℃. For wind speed, the fuzzy set has c=5m / s and d=2m / s, avoiding fuzziness jumps caused by instantaneous fluctuations in sea breeze. Real-time collected environmental data (such as 28℃, 65%RH, 4m / s) is calculated by a function, outputting the membership vectors of each fuzzy set (temperature: [0,0.47,0]; humidity: [0,0.67,0]; wind speed: [0,1,0]), completing the conversion from precise values ​​to fuzzy information.

[0060] Furthermore, a rule base was constructed using a dual-dimensional approach combining environmental coupling characteristics and material adaptability. The 27 rules comprehensively cover all combinations of 3 variables × 3 fuzzy sets. Each rule was validated using historical port data and reviewed and optimized by 5 senior engineers. The rule expressions adopt an IF premise (multivariate fuzzy combination) THEN conclusion (fuzzy set of rheological influence coefficients) structure. The conclusion categorizes the influence coefficients into three levels of fuzzy sets: small (0~0.3), medium (0.3~0.7), and large (0.7~1.0). Core rules include: IF high temperature AND humidity AND weak wind speed THEN large influence coefficient (suitable for scenarios where coal absorbs water and caking under high humidity and high temperature conditions); IF medium temperature AND dry humidity AND medium wind speed THEN small influence coefficient (corresponding to the optimal environment for routine port operations); IF low temperature AND humidity AND strong wind speed THEN medium influence coefficient (suitable for the slightly reduced ore fluidity under winter sea breeze conditions).

[0061] To improve the relevance of the rules, a dynamic correction mechanism based on material type is introduced: when dealing with highly viscous materials (such as bulk fertilizer), the conclusions of humidity-related rules are automatically upgraded by one level (e.g., from medium to large); when dealing with low-viscosity materials (such as iron ore), the conclusions of temperature-related rules are downgraded by one level, solving the problem of poor adaptability of traditional fixed rule libraries. The rule library uses a linked list structure for storage, supporting the online addition of rules for new scenarios (such as adding special rules for typhoon passage), ensuring scalability.

[0062] S412. Based on the rule base and fuzzy results, perform fuzzy reasoning, and defuzzify the reasoning results to obtain the environmental impact coefficient.

[0063] Based on the Mamdani inference method, a complete link of premise matching, intensity calculation, and conclusion synthesis is constructed to adapt to the real-time requirements of port environmental variables. First, premise matching is performed. By fuzzifying the membership vector of the output, rules that satisfy all premise conditions are selected (activation rules). For example, when the input is medium temperature (0.47), humidity (0.67), and medium wind speed (1), five association rules are activated, such as medium temperature + humidity + medium wind speed, medium temperature + humidity + weak wind speed. Then, the trigger intensity of each activation rule is calculated by AND operation (taking the minimum value of the membership degree of each premise variable). For example, the trigger intensity of the first activation rule is min(0.47,0.67,1)=0.47, ensuring that the rule is effective only when each premise reaches a certain degree of fuzziness. To address potential rule conflicts (such as two rules reaching contradictory conclusions), a priority mechanism is introduced: rules with a higher degree of matching the current material type take precedence over rules with a lower degree of matching (e.g., when processing grain, humidity-related rules have higher priority than temperature-related rules). Finally, the output is integrated through an OR operation (taking the maximum value of the fuzzy sets of each rule's conclusion). For example, if the rule trigger strengths in the two conclusions that are influence coefficients are 0.47 and 0.32 respectively, the membership degree of the fuzzy set after integration is taken as the maximum value of 0.47, forming the final output fuzzy set.

[0064] Furthermore, the universe of discourse of the output fuzzy set is determined to be [0,1], and it is discretized into 100 sampling points (step size 0.01). Each sampling point x corresponds to the membership degree μ(x) of the output fuzzy set. The centroid method is used to discretize and approximate the calculation to convert the fuzzy output into an accurate environmental impact coefficient. For example, if the output fuzzy set has μ=0.47 when x=0.4, μ=0.5 when x=0.5, and μ=0.47 when x=0.6, C≈0.5 can be obtained through calculation, that is, the environmental impact coefficient is 0.5. To further improve the accuracy, an outlier correction step is added: if the deviation of the calculation result from the average of the previous 5 times is >0.2 (such as a sudden increase in humidity due to a sudden rainstorm), the exponential smoothing method is used for correction (C correction = 0.7 × C current + 0.3 × C average) to avoid coefficient jumps caused by sudden changes in environmental parameters. The final output environmental impact coefficient is retained to two decimal places and strictly controlled within the range of [0,1]. The larger the value, the stronger the interference of the environment (such as high humidity agglomeration and low temperature solidification) on the rheological properties, providing a precise quantitative basis for subsequent external impact assessment.

[0065] Furthermore, the dynamic changes in the loading machine's operating parameters are analyzed to assess the intensity of its disturbance to rheological properties and quantify the stability of the operating state.

[0066] Specifically, extracting the operation status index includes the following steps: S421. Analyze the operating parameters and evaluate the disturbance intensity.

[0067] First, the core parameters are extracted: the conveyor belt speed is acquired in real time by an incremental encoder (accuracy ±0.1m / s, the number of output pulses is linearly related to the rotational speed v, and the pulse frequency f=60×v×Z / D, where Z is the number of encoder teeth and D is the diameter of the conveyor belt roller); the hopper opening is acquired by a lever-type displacement sensor (range 0-500mm, accuracy ±0.5mm, using 4-20mA analog signal output, which is converted into a digital quantity s through a signal conditioning module); and the robotic arm rotation angle is acquired by an absolute angle sensor (measurement range 0-360°, accuracy ±0.1°, achieving non-contact measurement based on the magnetoresistive effect).

[0068] Then, parameter cleaning was performed: a sliding window (window size of 5 sampling points) median filtering was used to remove pulse interference caused by mechanical vibration (such as instantaneous speed jumps when the conveyor belt starts and stops), outliers were identified by the adjacent data difference method and linear interpolation was used to complete them.

[0069] Finally, the parameter deviations are calculated as follows: Δv=|v-v0|, Δs=|s-s0|, Δθ=|θ-θ0| (Δθ is the deviation of the robotic arm angle). The preset optimal values ​​v0, s0, and θ0 are obtained through offline genetic algorithm optimization. With the goal of loading efficiency ≥200t / h and material blockage rate <1%, the stable loading data of the past 3 months are iteratively optimized. For example, for coal, v0=1.5m / s and s0=300mm, and for iron ore, v0=2.0m / s and s0=350mm, to ensure that the optimal values ​​are accurately matched with the material characteristics.

[0070] Furthermore, a perturbation assessment model is constructed using a piecewise exponential function and a material fit coefficient, which avoids oversensitivity to small deviations and accurately captures the impact of sudden changes in key parameters.

[0071] Specifically, when the parameter deviation is within the allowable fluctuation range (e.g., Δv ≤ 0.2 m / s), the disturbance intensity increases slowly; beyond this range, the intensity rises rapidly, meeting the requirements of tolerating small fluctuations in port loading and demanding emergency response to large sudden changes, thus satisfying:

[0072]

[0073] in, Indicates the intensity of the conveyor belt speed disturbance. This indicates the intensity of disturbance caused by the hopper opening. , , , This represents a proportionality coefficient, which is dynamically adjusted according to the material type. It is particularly useful when processing granular materials (such as iron ore). Increase by 30% (set to 13) because granular materials are more sensitive to sudden changes in conveyor belt speed, which can easily lead to accumulation; when processing powdery materials (such as cement), Increasing it by 40% (set to 0.14) is problematic because even a slight change in the hopper opening can cause dust overflow or blockage. A real-time correction mechanism is also introduced: when a shear rate peak > 300s is detected... -1 When materials are prone to clumping, the k-value of all parameters is uniformly increased by 20% to enhance sensitivity to operational fluctuations and proactively mitigate the risk of blockages. Robotic arm angle disturbance intensity. The formula is consistent with the hopper opening.

[0074] S422. Using the operating parameters and the disturbance intensity, synthesize the operating state index.

[0075] A random forest weighting method combined with dynamic adjustment is used to fuse the intensity of multiple perturbations, ensuring that the index accurately reflects the operational perturbation characteristics under different materials and operating conditions. First, the basic weights are determined through random forest model training: sample pairs are defined as operational perturbation intensity minus rheological parameter change (features are...). , , The label represents the actual rate of change of the rheological parameter (viscosity). 100 decision trees are trained to form a random forest, and the importance score of each perturbation intensity is output and normalized to the base weights. For example, the base weights are typically 0.4 for hopper opening, 0.35 for conveyor belt speed, and 0.25 for robotic arm angle. Because hopper opening directly controls material flow, the weight for high-viscosity materials (such as bulk fertilizer) needs to be increased by an additional 40% (to 0.56), while for low-viscosity materials (such as coal), the weight for conveyor belt speed is increased by 20% (to 0.42). Then, a weighted summation is performed to obtain the initial operating state index I_temp, followed by index calibration: a base rheological property index is introduced for correction. When the base index... When the index is >70 (high viscosity material), I = min(1.0, I_temp × 1.1) (amplifying the impact of disturbances); when the basic index is <30 (low viscosity material), I = max(0.0, I_temp × 0.9) (suppressing excessive quantization of small disturbances). The final output range is strictly controlled within the operating state index of 0-1. 0-0.3 is low disturbance (stable operation), 0.3-0.7 is medium disturbance (requires fine-tuning of parameters), and 0.7-1.0 is high disturbance (requires urgent optimization).

[0076] Furthermore, by coupling the environmental influence coefficient and the operating state index, the overall interference intensity of external factors on rheological properties is quantified.

[0077] Specifically, the analysis of external influencing factors based on environmental and operational impacts using environmental impact coefficients and operational status indices includes the following steps: S431. Construct a synergistic impact assessment model. Based on the environmental impact coefficient and the operational status index, calculate the initial external impact factor through the synergistic impact assessment model.

[0078] First, the original calculation records of the environmental impact coefficient E and the operational state index O are retrieved to verify the accuracy of the defuzzification centroid method calculation of E (e.g., whether outliers are corrected through exponential smoothing) and the dynamic weight allocation logic of O (e.g., whether it is combined with the basic rheological index calibration) to ensure the rigor of the standardization process. Then, a deviation threshold verification is used: if the deviation of E from the previous three mean values ​​is >0.15, or the deviation of O from the previous three mean values ​​is >0.2, it is determined to be a temporary outlier and replaced by linear interpolation (taking the average of the two valid data points before and after) to avoid interference from a single outlier in subsequent fusion. Finally, the verified standardized coefficients E' and O' are output. Both not only have consistent dimensions, but their data fluctuation range is also controlled within ±0.1 (stable operating conditions), providing high-quality input for the synergistic impact assessment.

[0079] Furthermore, a synergistic impact assessment model F=α×E'+β×O'+ρ×E'×O' is constructed using a coupling method of linear main effect and nonlinear synergistic effect. This model is then used to calculate external impact factors, fundamentally addressing the strong interference problem of port environment and operation where 1+1>2. Specifically, the weight coefficients are determined through two-dimensional sensitivity analysis: First, based on historical data, the fluctuation amplitude of rheological parameters when E and O change individually is calculated (single-factor sensitivity), resulting in a basic weight ratio of 40% for both E and O (α=0.4, β=0.4). Then, through orthogonal experimental design (E takes 0.2 / 0.5 / 0.8, O takes 0.2 / 0.5 / 0.8, a total of 9 combinations), the synergistic effect of the two is quantified. When E≥0.7 and O≥0.7, the material blockage rate increases by 65% ​​compared to the single-factor effect. Therefore, the weight of the synergistic term is set to ρ=0.2 to highlight the coupling effect in high-interference scenarios. The E'×O' term in the formula is the core of synergistic effect quantification. For example, in a high-humidity environment (E'=0.8) with a sudden change in hopper opening (O'=0.9), the synergistic term contributes 0.8×0.9×0.2=0.144, making F=0.4×0.8+0.4×0.9+0.144=0.804, which accurately reflects the superimposed risk of harsh environment + operational instability. In contrast, in a normal environment (E'=0.3) and under stable operation (O'=0.2), the synergistic term contributes only 0.012, and F=0.192, which is consistent with the actual working conditions with low interference.

[0080] S432. Correct the initial external influence factor based on the rheological trend prediction results to obtain an external influence factor based on environmental and operational influences.

[0081] A trend-correction coefficient mapping relationship is constructed by introducing rheological trend prediction results to achieve dynamic calibration of interference factors and avoid misjudgments caused by static correction. Specifically, based on the urgency of the trend classification: if the prediction shows a sharp fluctuation (e.g., a sudden change in viscosity prediction value of >15% within 5 seconds), it indicates that external interference has triggered a sudden risk of rheological property change, and the factor needs to be amplified to trigger high-intensity adjustment, so it is multiplied by a correction coefficient of 1.2; if the prediction shows an upward trend (slope >0.02 / second) and confidence level ≥0.8, it is multiplied by a correction coefficient of 1.1 to moderately strengthen the interference assessment; if it shows a downward trend or a stable trend, the original F value is maintained to avoid over-adjustment.

[0082] After correction, range constraints must be applied: if F-correction > 1.0, the anchor value is forced to 1.0 (representing extreme disturbances); if F-correction < 0.1, the anchor value is set to 0.1 (to avoid ineffective adjustments triggered by minor disturbances). For example, under certain operating conditions E' = 0.6 and O' = 0.7, the initial F = 0.4 × 0.6 + 0.4 × 0.7 + 0.2 × 0.6 × 0.7 = 0.544. If a sharp fluctuation is predicted, then F-correction = 0.544 × 1.2 ≈ 0.653, providing a precise basis for emergency adjustments.

[0083] S5. Based on the rheological trend, the comprehensive rheological state index, and the external influencing factors, the operating parameters are optimized in real time through reinforcement learning to achieve adaptive adjustment of rheological properties.

[0084] The analysis of the rheological trend includes the following steps: S511. Preprocess the historical rheological data of the port.

[0085] First, outlier stratification is implemented: physical constraints are used for screening (e.g., viscosity values ​​exceeding the common port material range of 0.01~100 Pa·s are directly identified as outliers). Then, the 3σ criterion is used to handle statistical outliers. Material type is used as the grouping unit (e.g., coal group, iron ore group). The mean μ and standard deviation σ of the rheological parameters for each group are calculated, and data with |x-μ|>3σ are removed to avoid interference from outliers across material types.

[0086] Then, time series alignment is performed: using the PLC master clock of the loading machine (1ms accuracy) as the reference, the scattered rheological parameters, environmental coefficients and operating index data are aligned according to the timestamp. For missing sampling points (such as 1 to 2 data gaps caused by temporary sensor failures), linear interpolation is used to fill in the gaps. If there are more than 5 consecutive gaps, they are marked as invalid time periods and removed to ensure the continuity of time series data (the time interval is stable at 100ms).

[0087] S512. Construct an online learning real-time update model based on an online learning algorithm, and predict the trend of rheological characteristics in the future from historical rheological data through the online learning real-time update model.

[0088] Incremental Support Vector Regression (ISVR) is used as the basic model. The initialization phase requires three steps: data screening, model structure design, and parameter optimization.

[0089] First, the initial training set is selected: stable operating condition data from the previous week are extracted from historical data. The data must meet the following requirements: operating condition index < 0.3, environmental impact coefficient fluctuation < 0.1, and each data point must be associated with clear material batch information (such as mine number and moisture content range) to ensure that the initial sample can reflect the rheological patterns under normal operation.

[0090] Model structure design: The input layer consists of 4-dimensional features (environmental impact coefficient E, operating state index O, current shear rate γ, and rheological parameter values ​​from the previous second), and the output layer consists of predicted rheological parameter values ​​for the next second. The radial basis function (RBF) is selected as the kernel function because it can effectively fit the nonlinear relationship of port material rheology. The kernel function parameter σ is determined by the grid search method.

[0091] Cross-validation is used for parameter optimization: the initial training set is divided into training and validation sets in a 7:3 ratio. The root mean square error (RMSE) is used as the objective function. The penalty coefficients C (range 1~100) and σ are optimized by particle swarm optimization (PSO). The initialized ISVR model can directly output continuous predicted values ​​of rheological parameters, providing a basis for subsequent trend judgment.

[0092] Furthermore, an incremental update-stability verification-anomaly backtracking mechanism is designed to ensure that the model can quickly adapt to the dynamic changes of port materials (such as batch switching and gradual changes in humidity) while avoiding noise interference introduced by new data. The update trigger condition is set at every 100 new data points collected (corresponding to a duration of 10 seconds, balancing real-time performance and sample representativeness). The update process consists of three steps: The first step is new data preprocessing, which involves outlier removal and normalization of the 100 data points, marking high-value samples (such as data on sudden changes in rheological parameters or operating condition indices > 0.7). These samples receive a 20% weight increase during the update. The second step is incremental learning execution, where the ISVR model calculates Lagrange multipliers only for the newly added samples and selects new support vectors (approximately 15% to 20% of the new samples) using KKT conditions to replace old support vectors with low contribution in the model. The third step is stability verification, where the updated model is used to re-predict the first 50 historical data points. If the prediction error increases by more than 20% compared to before the update, a partial rollback is triggered, retaining 50% of the old support vectors and fusing only 50% of the new support vectors to avoid model drift due to short-term abnormal data (such as sudden environmental changes caused by heavy rain).

[0093] Furthermore, based on the logic of continuous value prediction, trend classification, and confidence quantification, the system outputs a trend that is both accurate and readable, providing a clear basis for advance adjustments.

[0094] First, the prediction time domain is determined to be the next 5 seconds. This duration covers the response delay of the loading machine's operating parameters (about 2 to 3 seconds) and avoids the accuracy decay of long-term prediction. The prediction inputs are the current environmental influence coefficient E, the operating state index O, and the rheological parameter sequence of the past 3 seconds. The updated ISVR model continuously outputs the predicted rheological parameter values ​​every 100ms within the next 5 seconds (a total of 50 consecutive values).

[0095] Then, trend classification is performed: the linear fit slope k and fluctuation standard deviation σ_trend of 50 predicted values ​​are calculated. When the slope k > 0.02 / s and σ_trend < 0.05, it is determined to be an upward trend (such as the gradual increase of material viscosity in a high humidity environment); when k < -0.02 / s and σ_trend < 0.05, it is a downward trend (such as the decrease of material shear rate after the conveyor belt accelerates); the rest are stable trends. At the same time, a subclassification of severe fluctuations is added for special scenarios (such as a sudden change in predicted value > 10%) to trigger a higher priority adjustment response. Finally, the confidence level is quantified: the confidence level P comprehensively considers three factors: the model's recent prediction error (accounting for 50%, the smaller the error, the higher the P), the similarity between the current working conditions and the training set (accounting for 30%, if the similarity is >85%, P is increased by 0.2), and the stability of the environment and operating parameters (accounting for 20%, if the fluctuation is <0.1, P is increased by 0.1). The calculation range of P is strictly controlled within 0~1. The final output is a structured result of trend type + confidence level (e.g., upward trend, confidence level 0.85).

[0096] Furthermore, by combining rheological trend prediction, comprehensive state index, and external influencing factors, the operating parameters are optimized in real time through reinforcement learning to achieve adaptive adjustment of rheological properties.

[0097] Specifically, the adaptive adjustment of rheological properties by optimizing operating parameters in real time through reinforcement learning based on rheological trends, the comprehensive rheological state index, and the external influencing factors includes the following steps: S521. Based on the rheological trend, the comprehensive rheological state index, and the external influencing factors, define the state space and design the action space and reward function.

[0098] A high-dimensional state space is constructed by trend quantification and exponential interval discretization to ensure that the input can accurately map the dynamic flow scenario of port materials. The state vector is defined as (trend code, S interval code, F interval code) with a dimension of 3×1. Each component is combined with the output of the previous steps for engineering processing. First, the rheological trends are classified and coded: a stable trend is coded as 0, an upward trend as 1, a downward trend as 2, and a drastic fluctuation as 3. At the same time, the trend confidence level is introduced as an implicit weight (when the confidence level is ≥0.8, the weight of this component is increased to 1.2). Second, the comprehensive rheological state index S(0-1) is discretized into 5 intervals according to the risk level and coded: [0,0.3) (safe) = 0, [0.3,0.5) (low warning) = 1, [0.5,0.7) (medium warning) = 2, [0.7,0.9) (high warning) = 3, [0.9,1] (dangerous) = 4. Finally, the external influencing factor F(0-1) is discretized into 3 intervals and coded: [0,0.3) (low interference) = 0, [0.3,0.7) (medium interference) = 1, [0.7,1] (high interference) = 2. For example, under a certain working condition, step 5 outputs an upward trend with a confidence level of 0.85, step 7 outputs S=0.8, and step 8 outputs F=0.6. Then the state vector is (1,3,1), which retains the core information while reducing the computational complexity of the model and adapts to the needs of real-time decision-making.

[0099] Furthermore, based on the mechanical performance constraints of the loading machine and the material characteristics, a motion space with core parameters prioritized and adjustment ranges tiered is constructed, covering three key operating parameters. The adjustment logic of each parameter is associated with the optimal value (v0, s0, θ0) of the loading machine's operating parameters. The core parameters include: 1. Conveyor belt speed, the adjustment base is v0, the adjustment range is ±15% (the maximum mechanical speed adjustment range is ±20%, with a 5% safety redundancy reserved), and it is discrete in 5 levels of adjustment: -15% (rapid drop), -8% (slow drop), 0 (maintain), +8% (slow rise), +15% (rapid rise). When dealing with highly viscous materials, the maximum adjustment range is narrowed to ±10% to avoid material accumulation. 2. Hopper opening, the adjustment base is s0, the adjustment range is ±20% (limited by the opening sensor range), and it is discrete in 5 levels: -20% (significantly close), -10% (slightly close), 0 (maintain), +10% (slightly open), +20% (significantly open). When dealing with powdery materials (such as cement), the adjustment range is halved to ±10% to prevent dust from overflowing. 3. The rotation angle of the robotic arm is adjusted based on θ0, with an adjustment range of ±5° (robotic arm joint limit), and is discrete in 3 levels: -5° (left deviation), 0 (maintain), and +5° (right deviation). It is only activated when material off-center loading is detected (velocity field center of gravity shift > 10%).

[0100] The motion space contains 5×5×3=75 possible combinations of motions. Invalid combinations can be eliminated through parameter mutual exclusion constraints (such as a sudden drop of the conveyor belt + a large opening of the hopper, which is prone to blockage and is directly marked as an invalid motion), thereby improving decision-making efficiency.

[0101] Furthermore, a nonlinear reward function with core objective orientation and multiple constraint penalties is constructed, satisfying: R = 15 × (1-S) × P_trend - 8 × F - sum of penalty terms, where P_trend is the trend confidence level, ensuring a strong correlation between the reward and the core objectives of port operations (reducing congestion risk and stabilizing material flow). The design logic of each part is as follows: 1. The core reward item is 15×(1-S)×P_trend. The smaller S is (the more stable the state) and the higher P_trend is (the more reliable the trend), the more generous the reward. For example, when S=0.2 (safe) and P_trend=0.9, the reward for this item is 15×0.8×0.9=10.8 points; when S=0.9 (dangerous) and P_trend=0.85, the reward for this item is 15×0.1×0.85=1.28 points, highlighting the need for improvement in dangerous states.

[0102] 2. The interference penalty term is 8×F. The larger F is (the stronger the external interference), the heavier the penalty. For example, when F=0.8 (high interference), the penalty is 6.4 points, which guides the model to prioritize avoiding operational errors in high interference scenarios.

[0103] 3. The penalty items are further divided into three categories: penalty for excessive adjustment range (3 points are deducted when the adjustment of a single parameter is >10%, and 5 points are deducted when it is >15%, to avoid mechanical impact), penalty for parameter conflict (4 points are deducted for a combination of hopper opening and speed reduction, to prevent material accumulation), and penalty for frequent adjustment (2 points are deducted for performing the same adjustment in 2 consecutive steps, to reduce equipment wear and tear).

[0104] The reward value is controlled within the range of [-15, 15]. When R > 10, it is determined to be the optimal action, and when R < -5, it is determined to be the invalid action, providing clear guidance for model learning.

[0105] S522. Based on the state space, the action space, and the reward function, the operating parameters are optimized in real time through reinforcement learning to achieve adaptive adjustment of rheological characteristics.

[0106] An intelligent agent is built using a dual DQN + priority experience replay architecture to adapt to the dynamic nature of port operating conditions.

[0107] Offline training phase: 1. Dataset construction: Collect state-action-reward samples of multiple materials (coal, iron ore, grain) at the port over 12 months, and divide them into 3 subsets according to material type to ensure model generalization. 2. Network structure design: The input layer is a 3-dimensional state vector, the hidden layer has 2 layers (64 and 32 neurons), the activation function is ReLU, and the output layer is a 50-dimensional action value vector (corresponding to 50 valid actions). 3. Hyperparameter settings: initial learning rate is 0.001, decreasing to 0.0001 every 100,000 training steps; batch size is 32; experience replay pool capacity is 1 million records; priority is based on |TD error|. 0.6 Allocation prioritizes the learning of high-value samples (such as effective regulatory actions in dangerous situations).

[0108] Online execution phase: 1. Decision-making process: The state vector is read every 20ms (synchronized with the loading machine response cycle). An action is selected through an ε-greedy strategy (ε is initially 0.1 and decreases linearly to 0.01 over time to balance exploration and utilization). For example, in state (1,3,1) (upward trend, high warning, medium interference), the optimal action output is hopper opening -10% + conveyor belt speed +8%. 2. Real-time updates: Within 100ms after an action is executed, the new state (S_new, F_new) and reward R are obtained, the TD error is calculated and stored in the experience pool, the target network parameters are updated every 5 steps, and the main network parameters are updated every 20 steps to ensure the timeliness of learning; 3. Anomaly handling: When a shear rate peak > 400s is detected in step 2. -1 In case of emergency blockage risk, skip the DQN decision and directly execute the preset emergency action (closing the hopper by 20% + increasing the speed by 15%), with a response time of ≤50ms, to ensure operational safety.

[0109] Furthermore, an adaptive mechanism of reward feedback, parameter expansion, and scene memory is constructed to adapt to the characteristic changes of new material batches.

[0110] First, the basic optimization logic triggers the adjustment range expansion when the reward R < -3 for 3 consecutive steps (adjustment is ineffective) or R < 2 for 5 consecutive steps (adjustment effect is weak): the conveyor belt speed adjustment range is expanded from ±15% to ±20%, the hopper opening is expanded from ±20% to ±25% (while still retaining mechanical safety redundancy), and a ±8° adjustment level for the robotic arm angle is added to expand the motion space. Secondly, the range rollback mechanism gradually shrinks the range back to the original specification when R>8 occurs twice after expansion (adjustment is effective) to avoid over-adjustment. Next, for material batch adaptation, a scene memory factor is introduced. When a new material batch is detected (the change in the basic index is >20), the adjustment parameter range of historical similar batches is automatically called (e.g., if the similarity between a new coal batch and a batch from 3 months ago is >80%, then its adjustment range is reused), and the state-action samples of the previous 100 steps are marked as high priority to accelerate model adaptation. Finally, the manual intervention interface automatically triggers an audible and visual alarm and pauses adjustment when the comprehensive index S > 0.95 (extreme danger), waiting for operator confirmation before resuming, ensuring safety and controllability.

[0111] Please see Figure 2 In an embodiment, to efficiently execute the adaptive adjustment method for material rheological characteristics of a port intelligent loading machine provided by the present invention, the present invention also provides an adaptive adjustment system for material rheological characteristics of a port intelligent loading machine, comprising: an input device, an output device, a processor, and a memory, wherein the input device, output device, processor, and memory are interconnected, and the memory contains program instructions for the steps of the adaptive adjustment method for material rheological characteristics of a port intelligent loading machine. The adaptive adjustment system for material rheological characteristics of a port intelligent loading machine of the present invention has a compact structure and stable performance, and can stably execute the adaptive adjustment method for material rheological characteristics of a port intelligent loading machine of the present invention, further improving the overall applicability and practical application capability of the present invention.

[0112] In this embodiment, the processor may be a central processing unit, but it can also be other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. Input devices can be used to acquire data. Output devices can be used to output the results obtained by storing program instructions contained in a computer program in the memory provided by this invention. The memory may include read-only memory and random access memory (RAM), and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory (RAM).

[0113] In one possible implementation, the memory may include a stored program area and a stored data area. The stored program area may store the operating system and applications required for at least one function; the stored data area may store data created during use. Furthermore, the memory may include read-only memory and random access memory, and provides instructions and data to the processor. The memory stores the operating system and operating instructions, executable modules, or data structures, or subsets thereof, or extended sets thereof. The operating instructions may include various operation instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and handling hardware-based tasks.

[0114] The embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described adaptive adjustment method for the material rheological characteristics of the intelligent port loading machine.

[0115] The storage medium can include various media that can store program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0116] In summary, this invention correlates the physical properties and rheological characteristics of materials to obtain a basic index, anchoring the inherent rheological properties of materials and providing a benchmark for subsequent adjustments, thus solving the problem of blind adjustments caused by differences in the characteristics of different batches of materials. It utilizes optical flow methods to quantify the velocity field and shear rate to obtain dynamic characteristics, capturing the material's flow dynamics in real time, compensating for the shortcomings of static properties in reflecting changes in operating conditions, and providing data support for accurate judgment of rheological dynamics. By fusing multiple features through neural networks to obtain a comprehensive index, it integrates static basics, dynamic mechanics, and real-time parameters to generate a unified rheological state assessment index, realizing a shift from fragmented perception to holistic judgment, and accurately identifying safe / warning / dangerous states. Based on the analysis of external influencing factors of the environment and operation, it quantifies the comprehensive effect of external disturbances on rheology, avoiding adjustment deviations caused by isolated evaluation of single factors, making the adjustment strategy more aligned with actual operating conditions. Through reinforcement learning to optimize operating parameters, it dynamically adjusts based on rheological trends and comprehensive evaluation results, achieving predictive optimization, solving the problem of traditional adjustment lag, reducing material blockage rates, and improving loading efficiency and operational safety.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention.

Claims

1. A method for adaptive adjustment of material rheological characteristics of a port intelligent loading machine, characterized in that, Includes the following steps: By associating the physical properties and rheological characteristics of the materials, the basic rheological property index is obtained; Dynamic mechanical characteristics are obtained by quantifying the velocity field and shear rate using optical flow methods. A comprehensive rheological state index is obtained by fusing the current rheological parameters, the basic rheological property index, and the dynamic mechanical characteristics using a neural network. Using the environmental impact coefficient and operational status index, we analyze the external influencing factors based on environmental and operational impacts. Based on the rheological trend, the comprehensive rheological state index, and the external influencing factors, the operating parameters are optimized in real time through reinforcement learning to achieve adaptive adjustment of rheological properties. The process of obtaining a comprehensive rheological state index by fusing current rheological parameters, the basic rheological property index, and the dynamic mechanical characteristics using a neural network includes the following steps: Standardize the current rheological parameters, the basic rheological property index, and the dynamic mechanical characteristics; A fully connected neural network structure is constructed, and a comprehensive rheological state index is obtained through the fully connected neural network structure based on the standardized results. Obtaining the current rheological parameters includes the following steps: Calculate the shear stress; Based on the triple selection mechanism of basic index classification, material type adaptation, and dynamic working condition correction, the constitutive model is selected. By combining the shear stress and the constitutive model, the current rheological parameters are solved; The method of adaptively adjusting rheological properties by optimizing operating parameters in real time through reinforcement learning based on rheological trends, the comprehensive rheological state index, and the external influencing factors includes the following steps: Based on the rheological trend, the comprehensive rheological state index, and the external influencing factors, define the state space and design the action space and reward function; Based on the state space, the action space, and the reward function, the operating parameters are optimized in real time through reinforcement learning to achieve adaptive adjustment of rheological characteristics. The operating parameters include conveyor belt speed, hopper opening, and robotic arm rotation angle.

2. The adaptive adjustment method for material rheological characteristics of a port intelligent loading machine according to claim 1, characterized in that, The physical properties and rheological characteristics of the associated materials are used to obtain the basic rheological property index, including the following steps: Acquire sensor data and preprocess the data; Key features are extracted from preprocessed stationary data using a combination of statistical analysis and feature engineering. The dynamic weighted fusion algorithm is used to effectively integrate multiple features, and the basic rheological property index is calculated using the fused comprehensive feature vector.

3. The adaptive adjustment method for material rheological characteristics of a port intelligent loading machine according to claim 1, characterized in that, The method of quantifying the velocity field and shear rate using optical flow methods to obtain dynamic mechanical characteristics includes the following steps: Based on material flow data, calculate the optical flow field and shear rate distribution; Core features are extracted from high-dimensional velocity field and shear rate data to obtain dynamic mechanical characteristics.

4. The adaptive adjustment method for material rheological characteristics of a port intelligent loading machine according to claim 1, characterized in that, The analysis of external influencing factors based on environmental and operational impacts using environmental impact coefficients and operational status indices includes the following steps: A synergistic impact assessment model is constructed, and the initial external impact factors are calculated based on the environmental impact coefficient and the operational status index. The initial external influence factors are corrected by the rheological trend prediction results to obtain external influence factors based on environmental and operational influences.

5. The adaptive adjustment method for material rheological characteristics of a port intelligent loading machine according to claim 4, characterized in that, The extraction of the environmental impact coefficient includes the following steps: Fuzzyize environment variables and build a rule base; Fuzzy reasoning is performed based on the rule base and fuzzy results, and the reasoning results are defuzzified to obtain the environmental impact coefficient. Extracting the operational status index includes the following steps: Analyze the operating parameters and assess the disturbance intensity; The operating state index is synthesized using the operating parameters and the disturbance intensity.

6. The adaptive adjustment method for material rheological characteristics of a port intelligent loading machine according to claim 1, characterized in that, The analysis of the rheological trend includes the following steps: Preprocessing of historical rheological data of the port; An online learning real-time update model is constructed based on an online learning algorithm. This model is used to predict the future trend of rheological properties from historical rheological data.

7. An adaptive adjustment system for the material rheological characteristics of a port intelligent loading machine, characterized in that, The port intelligent loading machine material rheological characteristic adaptive adjustment system includes: an input device, an output device, a processor, and a memory. The input device, output device, processor, and memory are interconnected. The memory includes program instructions, which are used to execute the port intelligent loading machine material rheological characteristic adaptive adjustment method according to any one of claims 1-6.

Images