An ai optimization-based oil press multi-modal feeding data analysis and decision system

Abstract

The present application relates to the technical field of data processing, and more particularly to a kind of AI optimization-based oil press multi-modal feeding data analysis and decision system, comprising: data acquisition module, including data acquisition unit for collecting multi-modal feeding data of oil press;Data processing module, including pre-processing unit for pre-processing feeding data to output feeding features;Analysis decision module, including analysis unit for analyzing the feeding data according to machine learning model to output analysis result;Noise segmentation module, to determine adaptive noise frequency band segmentation threshold according to the decay rate of residual noise in feeding features per unit time;Transmission buffer module, to determine the adaptive buffer depth of feeding data transmission fluctuation according to the data point matching failure rate of feeding features;Window adjustment module, to determine the overlap ratio of adjacent sliding windows according to the prediction accuracy of time sequence label in analysis result.The present application improves the decision accuracy of feeding data.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an AI-optimized multimodal feeding data analysis and decision-making system for oil presses. Background Technology

[0002] In existing technologies, the feeding stage, as the first critical step in the oil pressing process, directly impacts subsequent pressing efficiency, oil yield, and product quality due to its stability, raw material quality compatibility, and equipment synergy. Traditional oil presses rely heavily on manual experience or simple automated control for the feeding process. Therefore, given the complexity and high dynamism of the feeding process, there is an urgent need to construct a decision-making system integrating multimodal data acquisition, intelligent noise processing, dynamic transmission adaptation, and AI deep analysis. Through precise perception, intelligent analysis, and adaptive optimization, this system can achieve high efficiency, stability, and low consumption in the feeding process, providing core support for improving the quality and efficiency of oil pressing production.

[0003] Chinese Patent Publication No. CN115952900A discloses a smart energy operation optimization method based on AI analysis, including: (1) constructing an optimization algorithm model based on historical data, utilizing historical power plant operation data, fully considering external boundary conditions (such as atmospheric pressure, ambient temperature, humidity, etc.), setting different evaluation indicators according to different optimization parameters, reasonably classifying the operation under various operating conditions (unit load), finding the optimal operation of the unit (system, equipment), and displaying the optimal operation online; (2) constructing a real-time calculation model for equipment (system) performance indicators, realizing real-time minute-level and hour-level calculation of equipment (system) performance indicators. According to real-time online calculation and analysis, the current operating status and historical best status of the equipment (system) are displayed in real time. The calculation data results guide the operation management in real time; (3) Construct an online self-learning model: realize the online comparison of real-time operating conditions, equipment operating indicators and historical best operating conditions, continuously record the best operating conditions, conduct the best comparison, deeply explore the best operating conditions, and continuously improve the energy-saving and efficiency-enhancing level of unit operation; (4) Based on the correlation analysis of optimization parameters, sort the parameters that affect the correlation of optimization parameters from high to low, and through the comparison of the current value and the historical best value, refine and optimize the optimization parameters, and continuously optimize and adjust them towards the ultimate goal of the best operating conditions. It can be seen that the AI-based smart energy operation optimization method has the problem that the model learns noise patterns rather than real rules due to the lack of effective noise filtering, and cannot accurately distinguish between effective signals and noise, thus causing a decrease in the decision accuracy of material feeding data. Summary of the Invention

[0004] To address this issue, the present invention provides an AI-optimized multimodal feeding data analysis and decision-making system for oil presses, which overcomes the problem in existing technologies where ineffective noise filtering leads to the model learning noise patterns rather than real patterns, making it impossible to accurately distinguish between valid signals and noise, thus resulting in a decrease in the accuracy of feeding data decision-making.

[0005] To achieve the above objectives, this invention provides an AI-optimized multimodal feeding data analysis and decision-making system for oil presses, comprising:

[0006] The data acquisition module includes a data acquisition unit for acquiring multimodal feeding data of the oil press and a data transmission unit connected to the data acquisition unit for transmitting the feeding data to the processing position;

[0007] The data processing module, which is connected to the data acquisition module, includes a preprocessing unit for preprocessing the feeding data to output feeding characteristics and a model training unit connected to the preprocessing unit for training an initial model based on the feeding characteristics to output a machine learning model.

[0008] An analysis and decision-making module, which is connected to the data processing module, includes an analysis unit that analyzes the feeding data according to the machine learning model to output analysis results, and a decision-making unit connected to the analysis unit to make decisions based on the analysis results to optimize the feeding process of the oil press.

[0009] A noise segmentation module, which is connected to the data processing module, is used to determine the adaptive noise band segmentation threshold based on the attenuation rate of residual noise in the feeding characteristics per unit time.

[0010] A transmission buffer module, which is connected to the data acquisition module and the noise segmentation module respectively, is used to determine the adaptive buffer depth of the feeding data transmission fluctuation based on the data point matching failure rate of the feeding characteristics.

[0011] A window adjustment module, which is connected to the data processing module and the transmission buffer module respectively, is used to determine the overlap ratio of adjacent sliding windows based on the prediction accuracy of the time series labels in the analysis results.

[0012] Furthermore, the noise segmentation module responds to the fact that the attenuation rate of the residual noise in the feeding characteristics is greater than or equal to a preset second attenuation rate, thus determining that the decision accuracy of the feeding data meets the requirements.

[0013] The noise segmentation module responds to the fact that the attenuation rate of the residual noise in the feeding feature is less than the preset second attenuation rate, and the accuracy of the decision to determine the feeding time does not meet the requirements.

[0014] Furthermore, the noise segmentation module responds to the fact that the attenuation rate of the residual noise in the feeding feature is greater than the preset first attenuation rate and less than the preset second attenuation rate, and preliminarily determines that the synergy of multimodal fusion does not meet the requirements.

[0015] Furthermore, the noise segmentation module adaptively adjusts the noise frequency band segmentation threshold in response to the attenuation rate of the residual noise in the feeding feature being less than or equal to the preset first attenuation rate.

[0016] Specifically, when the intensity of high-frequency noise increases and the dominant frequency rises, the noise band segmentation threshold is increased; when the dominant noise frequency decreases, the noise band segmentation threshold is decreased.

[0017] Furthermore, if the failure rate of data point matching in response to the feeding feature of the transmission buffer module is less than or equal to a preset first failure rate, it is determined that the synergy of multimodal fusion meets the requirements.

[0018] The transmission buffer module determines that the multimodal fusion coordination does not meet the requirements when the data point matching failure rate of the feeding feature is greater than the preset first failure rate.

[0019] Furthermore, in response to the data point matching failure rate of the feeding feature being greater than the preset first failure rate and less than or equal to the preset second failure rate, the transmission buffer module increases the adaptive buffer depth for feeding data transmission fluctuations.

[0020] Furthermore, the transmission buffer module responds to the fact that the data point matching failure rate of the feeding feature is greater than the preset second failure rate, and preliminarily determines that the fusion and model adaptability do not meet the requirements.

[0021] Furthermore, the increase in the adaptive buffer depth of the feeding data transmission fluctuation is determined by the difference between the data point matching failure rate of the feeding feature and the preset first failure rate.

[0022] Furthermore, the window adjustment module determines that the fusion and model fit meet the requirements when the prediction accuracy of the time series labels in the analysis results is greater than or equal to the preset accuracy.

[0023] The window adjustment module responds to the fact that the prediction accuracy of the time series labels in the analysis results is less than the preset accuracy, determines that the fusion and model adaptability do not meet the requirements, and increases the overlap ratio of adjacent sliding windows.

[0024] Furthermore, the increase in the overlap ratio of adjacent sliding windows is determined by the difference between the preset accuracy and the prediction accuracy of the time series labels in the analysis results.

[0025] Compared with existing technologies, the beneficial effects of this invention are as follows: The system of this invention, by setting up a data acquisition module, a data processing module, an analysis and decision-making module, a noise segmentation module, a transmission buffer module, and a window adjustment module, adjusts the adaptive noise frequency band segmentation threshold according to the attenuation rate of residual noise in the feeding characteristics per unit time. Since mechanical vibration during the feeding process causes high-frequency noise in the sensor data, and the noise intensity may change with operating conditions, fixed preprocessing algorithm parameters lead to inaccurate model learning. By adaptively adjusting the noise frequency band segmentation threshold, the threshold can always match the high-frequency boundary of the current noise, ensuring that effective signals are not falsely filtered out, while accurately covering the noise frequency band. The adaptive buffer depth of the feeding data transmission fluctuation is adjusted according to the data point matching failure rate of the feeding characteristics. Because different sensors have different acquisition frequencies and delays, time alignment is not performed during fusion, leading to information mismatch across time points. By increasing the feeding data transmission fluctuation... The adaptive buffer depth allows for more space to temporarily store early-arriving sensor data, waiting for data from sensors with higher latency or lower frequency to complete the data. This provides a more complete time window and data foundation for the time alignment algorithm, reducing mismatches caused by data gaps or overshoot. The overlap ratio of adjacent sliding windows is adjusted based on the prediction accuracy of the time series labels in the analysis results. Since the fusion strategy uses time series sliding window fusion and the model uses a static CNN, the model cannot learn the temporal sequence relationship within the window (such as the causal relationship of increased impurities followed by decreased flow). It can only treat the time series data as static features, leading to a disconnect between the temporal assumptions of the fusion strategy and the model's capabilities. By increasing the overlap ratio of adjacent sliding windows, more time steps can be shared between adjacent windows. In the prediction of continuous windows, the model can indirectly perceive the temporal correlation through the feature association of the overlapping parts, thereby alleviating the disconnect between the fusion strategy and the model's capabilities and improving the decision accuracy of the material feeding data.

[0026] Furthermore, the system of the present invention adjusts the adaptive noise frequency band segmentation threshold by setting a preset first attenuation rate and a preset second attenuation rate. Since mechanical vibration during the feeding process causes high-frequency noise in the sensor data, and the noise intensity may change with the working conditions, the fixed preprocessing algorithm parameters lead to inaccurate model learning. By adaptively adjusting the noise frequency band segmentation threshold, the threshold can always match the high-frequency boundary of the current noise, ensuring that the effective signal is not mistakenly filtered out, while accurately covering the noise frequency band, further improving the decision accuracy of the feeding data.

[0027] Furthermore, the system described in this invention adjusts the adaptive buffer depth of the feeding data transmission fluctuation by setting a preset first failure rate and a preset second failure rate. Since different sensors have different acquisition frequencies and delays, they are not time-aligned during fusion, resulting in information mismatch across time points. By increasing the adaptive buffer depth of the feeding data transmission fluctuation, there is more space to temporarily store the sensor data that arrives earlier, waiting for the sensor data with higher delay or lower frequency to be supplemented. This provides a more complete time window and data foundation for the time alignment algorithm, reduces mismatch caused by data gaps or advances, and further improves the decision accuracy of the feeding data.

[0028] Furthermore, the system described in this invention adjusts the overlap ratio of adjacent sliding windows by setting a preset accuracy rate. Since the fusion strategy adopts temporal sliding window fusion and the model uses a static CNN, the model cannot learn the temporal relationship within the window (such as the causal relationship of increased impurities followed by decreased flow rate). It can only treat the temporal data as static features, which leads to a disconnect between the temporal assumptions of the fusion strategy and the model's capabilities. By increasing the overlap ratio of adjacent sliding windows, the adjacent windows can share more time steps. In the prediction of continuous windows, the model can indirectly perceive the temporal correlation through the feature association of the overlapping parts, thereby alleviating the disconnect between the fusion strategy and the model's capabilities and further improving the decision accuracy of the material feeding data. Attached Figure Description

[0029] Figure 1 This is a block diagram of the overall structure of the AI-optimized multimodal feeding data analysis and decision-making system for oil presses, as described in an embodiment of the present invention.

[0030] Figure 2 This is a flowchart illustrating the process of determining the adaptive noise frequency band segmentation threshold in the AI-optimized multimodal feeding data analysis and decision-making system for oil presses according to an embodiment of the present invention.

[0031] Figure 3 This is a flowchart illustrating the process of determining the adaptive buffer depth for data transmission fluctuations in an AI-optimized multimodal feeding data analysis and decision-making system for an oil press, as described in an embodiment of the present invention.

[0032] Figure 4 This is a flowchart illustrating the process of determining the overlap ratio of adjacent sliding windows in the AI-optimized multimodal feeding data analysis and decision-making system for oil presses, as described in an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0034] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0035] Please see Figure 1 , Figure 2 , Figure 3 as well as Figure 4 The diagrams shown are, respectively, the overall structural block diagram of the AI-optimized multimodal feeding data analysis and decision-making system for an oil press according to an embodiment of the present invention; the logical flowchart of the process for determining the adaptive noise frequency band segmentation threshold; the logical flowchart of the process for determining the adaptive buffer depth of feeding data transmission fluctuations; and the logical flowchart of the process for determining the overlap ratio of adjacent sliding windows. The present invention provides an AI-optimized multimodal feeding data analysis and decision-making system for an oil press, comprising:

[0036] The data acquisition module includes a data acquisition unit for acquiring multimodal feeding data of the oil press and a data transmission unit connected to the data acquisition unit for transmitting the feeding data to the processing position;

[0037] The data processing module, which is connected to the data acquisition module, includes a preprocessing unit for preprocessing the feeding data to output feeding characteristics and a model training unit connected to the preprocessing unit for training an initial model based on the feeding characteristics to output a machine learning model.

[0038] An analysis and decision-making module, which is connected to the data processing module, includes an analysis unit that analyzes the feeding data according to the machine learning model to output analysis results, and a decision-making unit connected to the analysis unit to make decisions based on the analysis results to optimize the feeding process of the oil press.

[0039] A noise segmentation module, which is connected to the data processing module, is used to determine the adaptive noise band segmentation threshold based on the attenuation rate of residual noise in the feeding characteristics per unit time.

[0040] A transmission buffer module, which is connected to the data acquisition module and the noise segmentation module respectively, is used to determine the adaptive buffer depth of the feeding data transmission fluctuation based on the data point matching failure rate of the feeding characteristics.

[0041] A window adjustment module, which is connected to the data processing module and the transmission buffer module respectively, is used to determine the overlap ratio of adjacent sliding windows based on the prediction accuracy of the time series labels in the analysis results.

[0042] Specifically, the feeding data includes the vibration amplitude of the oil press, the amount of material fed per unit time, and the flow rate of the raw materials.

[0043] Specifically, preprocessing includes cleaning, noise reduction, integration, and feature extraction.

[0044] Specifically, the feeding characteristics include the amount of material fed per unit time of the cleaned oil press, the flow rate of the integrated raw materials, and the vibration amplitude of the oil press after noise reduction.

[0045] Specifically, machine learning models can be random forests, temporal convolutional networks, or multilayer perceptrons.

[0046] Specifically, the analysis results include the raw material quality grade, the type of abnormality of the oil press, and the parameter adjustment scheme.

[0047] Specifically, decisions are made based on the analysis results to optimize the feeding process of the oil press. After locating the problem based on the analysis results, the corresponding parameters are adjusted to improve production efficiency and quality.

[0048] Specifically, the noise frequency band segmentation threshold is the critical value used in the signal preprocessing stage to distinguish between the noise frequency band and the effective signal frequency band. Its core function is to accurately separate noise from the mixed signal collected by the sensor, providing a clean and effective signal for subsequent feature extraction and model analysis.

[0049] Specifically, the adaptive buffer depth for fluctuations in data transmission during material feeding is a dynamically adjusted parameter used to balance data transmission stability and system resource efficiency. Its core function is to cope with rate fluctuations in multimodal data (such as vibration, image, current, flow, and other signals) during transmission, ensuring that data flows continuously and completely into the AI ​​analysis module.

[0050] Specifically, the overlap ratio of adjacent sliding windows is a key parameter used in time series data processing to balance the preservation of time series continuity and data processing efficiency. Its core meaning is the proportion of the overlap between two adjacent data windows (i.e., data segments within a certain time period) to the length of a single window when processing continuous feeding time series data (such as signals that change with time, such as flow rate, vibration, and impurity content) in blocks.

[0051] In implementation, the system of this invention, by setting up a data acquisition module, a data processing module, an analysis and decision-making module, a noise segmentation module, a transmission buffer module, and a window adjustment module, adjusts the adaptive noise frequency band segmentation threshold according to the attenuation rate of residual noise in the feeding characteristics per unit time. Because mechanical vibration during the feeding process causes high-frequency noise in the sensor data, and the noise intensity may change with operating conditions, fixed preprocessing algorithm parameters lead to inaccurate model learning. By adaptively adjusting the noise frequency band segmentation threshold, the threshold can always match the high-frequency boundary of the current noise, ensuring that effective signals are not falsely filtered out, while accurately covering the noise frequency band. The adaptive buffer depth of feeding data transmission fluctuations is adjusted according to the data point matching failure rate of the feeding characteristics. Because different sensors have different acquisition frequencies and delays, time alignment is not performed during fusion, leading to information mismatch across time points. Increasing the adaptive buffer depth of feeding data transmission fluctuations... The increased overlap allows for more space to temporarily store early-arriving sensor data, waiting for data from sensors with higher latency or lower frequency to complete the process. This provides a more complete time window and data foundation for the time alignment algorithm, reducing mismatches caused by data gaps or overshoot. The overlap ratio of adjacent sliding windows is adjusted based on the prediction accuracy of the time series labels in the analysis results. Since the fusion strategy uses time series sliding window fusion and the model uses a static CNN, the model cannot learn the temporal relationships within the window (such as the causal relationship of increased impurities followed by decreased flow). It can only treat the time series data as static features, leading to a disconnect between the temporal assumptions of the fusion strategy and the model's capabilities. By increasing the overlap ratio of adjacent sliding windows, more time steps can be shared between adjacent windows. In the prediction of continuous windows, the model can indirectly perceive the temporal correlation through the feature association of the overlapping parts, thereby alleviating the disconnect between the fusion strategy and the model's capabilities and improving the decision-making accuracy of the material loading data.

[0052] Specifically, the noise segmentation module responds to the attenuation rate of residual noise in the feeding characteristics being greater than or equal to a preset second attenuation rate, thus determining that the decision accuracy of the feeding data meets the requirements.

[0053] The noise segmentation module responds to the fact that the attenuation rate of the residual noise in the feeding feature is less than the preset second attenuation rate, and the accuracy of the decision to determine the feeding time does not meet the requirements.

[0054] Specifically, the noise segmentation module responds to the fact that the attenuation rate of the residual noise in the feeding feature is greater than the preset first attenuation rate and less than the preset second attenuation rate, initially determines that the synergy of multimodal fusion does not meet the requirements, and determines whether the synergy of multimodal fusion meets the requirements based on the data point matching failure rate of the feeding feature.

[0055] It is understandable that the preset first attenuation rate is less than the preset second attenuation rate, and the three intervals divided by the preset first attenuation rate and the preset second attenuation rate correspond to three different situations:

[0056] The first interval is when the attenuation rate of residual noise in the feeding feature is less than or equal to the preset first attenuation rate. The corresponding situation is: due to mechanical vibration during the feeding process, high-frequency noise appears in the sensor data, and the noise intensity may change with the working conditions. The preprocessing algorithm parameters are fixed, resulting in inaccurate model learning.

[0057] The second interval is where the attenuation rate of residual noise in the feeding feature is greater than the preset first attenuation rate and less than the preset second attenuation rate. The corresponding situation is: due to the differences in the acquisition frequency and delay of different sensors, time alignment was not performed during fusion, resulting in information mismatch across time points.

[0058] The third interval is when the attenuation rate of residual noise in the feeding characteristics is greater than or equal to the preset second attenuation rate, which corresponds to the situation where the decision accuracy of the feeding data meets the requirements.

[0059] Understandably, in the AI-optimized multimodal feeding data analysis and decision-making system for oil presses, the first and second attenuation rates characterize the decision-making accuracy of the feeding data. The core logic is to indirectly quantify the impact of data quality on decision-making through the attenuation characteristics of residual noise, while using tiered thresholds to achieve refined and operational judgment of decision accuracy. The first attenuation rate serves as the critical value for severely deteriorated data quality. When the attenuation rate is less than or equal to the first attenuation rate, noise has accumulated significantly (attenuation is too slow), affecting not only decision-making but also potentially causing the noise segmentation threshold to fail. The second attenuation rate serves as the critical value for acceptable decision accuracy. When the attenuation rate is greater than or equal to the second attenuation rate, noise interference is weak enough not to affect core decisions, directly indicating that accuracy meets requirements. The preset first and second attenuation rates can be set according to actual working conditions. The setting of the preset first and second attenuation rates aims to ensure the accuracy and practicality of multimodal feeding data decision-making for oil presses. Optionally, the preset first attenuation rate and the preset second attenuation rate are determined through a limited number of experiments by evaluating the effect of different noise conditions on the decision-making process of feeding data. The determined preset first attenuation rate and preset second attenuation rate should meet the requirement that they are neither too small nor cause excessive interference to the decision-making process of feeding time. For example, the preset first attenuation rate is generally selected in the range of [0.4% / s, 0.6% / s], and the preset second attenuation rate is generally selected in the range of [0.7% / s, 0.9% / s].

[0060] Preferably, the first attenuation rate is 0.5% / s in a preferred embodiment, and the second attenuation rate is 0.8% / s in a preferred embodiment.

[0061] Specifically, in this scheme, residual noise is mainly considered as the percentage of noise in the signal. Therefore, the unit of the attenuation rate of residual noise in the feeding characteristics is % / s.

[0062] Specifically, the attenuation rate of residual noise in the feeding characteristics is the ratio of the difference between the noise energy before preprocessing and the residual noise energy after preprocessing to the noise energy before preprocessing.

[0063] Specifically, the attenuation rate of residual noise in the feeding feature ranges from [0,1], and the closer the value is to 1, the better the suppression effect.

[0064] In practice, the system of the present invention determines the decision accuracy of feeding data by setting a preset first attenuation rate and a preset second attenuation rate, thereby reducing the impact of decreased decision stability of feeding data due to inaccurate determination of the decision accuracy of feeding data and further improving the decision accuracy of feeding data.

[0065] Specifically, the noise segmentation module adaptively adjusts the noise frequency band segmentation threshold in response to the attenuation rate of the residual noise in the feeding feature being less than or equal to the preset first attenuation rate.

[0066] Specifically, when the intensity of high-frequency noise increases and the dominant frequency rises, the noise band segmentation threshold is increased; when the dominant noise frequency decreases, the noise band segmentation threshold is decreased.

[0067] Specifically, the increase / decrease of the noise band segmentation threshold is determined by the difference between the preset first attenuation rate and the attenuation rate of the residual noise in the feeding feature.

[0068] Specifically, when the difference between the preset first attenuation rate and the attenuation rate of the residual noise in the feeding feature is within 0.2% / s, the noise band segmentation threshold is increased to 1.1 times the original value. When the difference between the preset first attenuation rate and the attenuation rate of the residual noise in the feeding feature exceeds 0.2% / s, the noise band segmentation threshold is increased by 5Hz for every 0.1% / s exceeding the original value, in addition to increasing to 1.1 times the original value. For example, if the difference between the preset first attenuation rate and the attenuation rate of the residual noise in the feeding feature is 0.4% / s, and the current noise band segmentation threshold is 50Hz, the increased noise band segmentation threshold is 50×1.1+5×2=65Hz.

[0069] Specifically, when the difference between the preset first attenuation rate and the attenuation rate of the residual noise in the feeding feature is within 0.3% / s, the noise band segmentation threshold is reduced to 0.9 times the original value. When the difference between the preset first attenuation rate and the attenuation rate of the residual noise in the feeding feature exceeds 0.3% / s, the noise band segmentation threshold is reduced by 3Hz for every 0.1% / s exceeding the original value, in addition to being reduced to 0.9 times the original value. For example, if the difference between the preset first attenuation rate and the attenuation rate of the residual noise in the feeding feature is 0.5% / s, and the current noise band segmentation threshold is 100Hz, the reduced noise band segmentation threshold is 100×0.9-3×2=84Hz.

[0070] In practice, the system of the present invention adjusts the adaptive noise frequency band segmentation threshold by setting a preset first attenuation rate and a preset second attenuation rate. Due to the mechanical vibration during the feeding process, high-frequency noise appears in the sensor data, and the noise intensity may change with the working conditions. The preprocessing algorithm parameters are fixed, which leads to inaccurate model learning. By adaptively adjusting the noise frequency band segmentation threshold, the threshold can always match the high-frequency boundary of the current noise, ensuring that the effective signal is not mistakenly filtered out, while accurately covering the noise frequency band, further improving the decision accuracy of the feeding data.

[0071] Specifically, if the failure rate of data point matching in response to the feeding feature of the transmission buffer module is less than or equal to a preset first failure rate, it is determined that the synergy of multimodal fusion meets the requirements.

[0072] The transmission buffer module determines that the multimodal fusion coordination does not meet the requirements when the data point matching failure rate of the feeding feature is greater than the preset first failure rate.

[0073] Specifically, the transmission buffer module increases the adaptive buffer depth of data transmission fluctuations in response to the data point matching failure rate of the feeding feature being greater than the preset first failure rate and less than or equal to the preset second failure rate.

[0074] Specifically, the transmission buffer module responds to the data point matching failure rate of the feeding feature being greater than the preset second failure rate, preliminarily determining that the fusion and model's adaptability does not meet the requirements, and then determines whether the fusion and model's adaptability meets the requirements based on the prediction accuracy of the time series labels in the analysis results.

[0075] It is understandable that the preset first failure rate is lower than the preset second failure rate. The three intervals divided by the preset first failure rate and the preset second failure rate correspond to three different scenarios:

[0076] The failure rate of matching data points with the material feeding feature in the first interval is less than or equal to the preset first failure rate. The corresponding situation is: the synergy of multimodal fusion is determined to meet the requirements.

[0077] The second interval is where the failure rate of data point matching for the material feeding feature is greater than the preset first failure rate and less than or equal to the preset second failure rate. The corresponding situation is: due to the differences in the acquisition frequency and delay of different sensors, time alignment was not performed during fusion, resulting in information mismatch across time points.

[0078] The third interval is where the failure rate of matching data points for the feeding feature is greater than the preset second failure rate. The corresponding situation is as follows: Since the fusion strategy adopts temporal sliding window fusion and the model uses static CNN, the model cannot learn the temporal relationship within the window (such as the causal relationship of increasing impurities first and then decreasing flow rate). It can only treat the temporal data as static features, which leads to the disconnect between the temporal assumptions of the fusion strategy and the model's capabilities.

[0079] Understandably, in an AI-optimized multimodal feeding data analysis and decision-making system for oil presses, the data point matching failure rate (the proportion of different modal data that cannot be correctly matched in terms of time, features, or space) is a core indicator for measuring the synergy of multimodal fusion. The better the synergy, the more complementary the data from different modalities (such as flow sensors, vibration sensors, and image recognition modules) can be in terms of time sequence, logic, or features, resulting in a lower matching failure rate. Setting a first failure rate and a second failure rate as tiered thresholds essentially uses a step-by-step quantitative standard to accurately characterize different states of multimodal fusion synergy and correspond to differentiated system response strategies. The first failure rate, as the synergy qualification line, primarily defines whether synergy meets the standard. The second failure rate, as the anomaly severity escalation line, distinguishes between minor, repairable anomalies and severe anomalies requiring in-depth investigation. The preset first and second failure rates can be set according to actual working conditions. The setting of the preset first and second failure rates aims to ensure the accuracy and practicality of multimodal feeding data decision-making for oil presses. Optionally, the preset first failure rate and preset second failure rate are determined through a limited number of trials by evaluating the decision-making effect of different matching failure scenarios on the material feeding data. The determined preset first failure rate and preset second failure rate should be neither too small nor too disruptive to the decision-making process of the material feeding data. For example, the preset first failure rate is generally selected in the range of [1%, 3%], and the preset second failure rate is generally selected in the range of [4%, 6%].

[0080] Preferably, the preset first failure rate is 2% in a preferred embodiment, and 5% in a preferred embodiment.

[0081] Specifically, the data point matching failure rate of the feeding feature is the ratio of the number of data point matching failures to the total number of matches.

[0082] In practice, the system of the present invention determines the synergy of multimodal fusion by setting a preset first failure rate and a preset second failure rate, thereby reducing the impact of inaccurate determination of the synergy of multimodal fusion leading to a decrease in the decision accuracy of material feeding data and further improving the decision accuracy of material feeding data.

[0083] Specifically, the increase in the adaptive buffer depth of the feeding data transmission fluctuation is determined by the difference between the data point matching failure rate of the feeding feature and the preset first failure rate.

[0084] Specifically, when the difference between the data point matching failure rate of the feeding feature and the preset first failure rate is within 2%, the adaptive buffer depth of the feeding data transmission fluctuation is increased to 1.2 times the original value. When the difference between the data point matching failure rate of the feeding feature and the preset first failure rate exceeds 2%, in addition to increasing to 1.2 times the original value, the adaptive buffer depth of the feeding data transmission fluctuation is increased by 10KB for every 1% exceeding the original value. For example, if the difference between the data point matching failure rate of the feeding feature and the preset first failure rate is 4%, and the current adaptive buffer depth of the feeding data transmission fluctuation is 1280KB, the increased adaptive buffer depth of the feeding data transmission fluctuation will be 1280×1.2+10×2=1556KB.

[0085] In implementation, the system of the present invention adjusts the adaptive buffer depth of the feeding data transmission fluctuation by setting a preset first failure rate and a preset second failure rate. Due to the differences in the acquisition frequency and delay of different sensors, time alignment is not performed during fusion, resulting in information mismatch across time points. By increasing the adaptive buffer depth of the feeding data transmission fluctuation, there is more space to temporarily store the sensor data that arrives earlier, waiting for the sensor data with higher delay or lower frequency to be supplemented. This provides a more complete time window and data foundation for the time alignment algorithm, reduces mismatch caused by data gaps or advances, and further improves the decision accuracy of feeding data.

[0086] Specifically, the window adjustment module determines that the fusion and model fits the requirements when the prediction accuracy of the time series labels in the analysis results is greater than or equal to the preset accuracy.

[0087] The window adjustment module responds to the fact that the prediction accuracy of the time series labels in the analysis results is less than the preset accuracy, determines that the fusion and model adaptability do not meet the requirements, and increases the overlap ratio of adjacent sliding windows.

[0088] It is understandable that the two intervals for the preset accuracy rate division correspond to two different scenarios:

[0089] The first interval is when the prediction accuracy of the time series labels in the analysis results is less than the preset accuracy. The corresponding situation is: because the fusion strategy adopts time series sliding window fusion and the model uses static CNN, the model cannot learn the temporal relationship within the window (such as the causal relationship of increasing impurities first and then decreasing flow). It can only treat the time series data as static features, which leads to the disconnect between the temporal assumptions of the fusion strategy and the model's capabilities.

[0090] The second interval is when the prediction accuracy of the time series labels in the analysis results is greater than or equal to the preset accuracy, which corresponds to the situation where the fusion and model fit meet the requirements.

[0091] Understandably, in an AI-optimized multimodal feeding data analysis and decision-making system for oil presses, the prediction accuracy of time-series labels is used to characterize the adaptability of the fusion and the model. The core reason is that accuracy directly and quantitatively reflects the degree of matching between the multimodal fusion results and the needs of the AI ​​model. Prediction accuracy is precisely the performance indicator of the model completing the task (time-series label prediction) based on the fused features. High accuracy indicates that the fused features precisely meet the model's information needs (e.g., time-series patterns and feature correlations are effectively preserved), meaning good adaptability. Low accuracy indicates that the fused features fail to provide the effective information required by the model (e.g., key features are lost, and time-series correlations are disrupted), meaning poor adaptability. The preset accuracy can be set according to actual working conditions. Setting the preset accuracy aims to ensure the accuracy and practicality of the multimodal feeding data decision-making for oil presses. Optionally, the preset accuracy is determined through a limited number of experiments by evaluating the decision-making effect of different prediction accuracies on the feeding data. The determined preset accuracy should be neither too low nor too high, and should not cause excessive interference to the feeding data decision-making process. For example, the preset accuracy rate is typically selected in the range of [93%, 97%].

[0092] Preferably, the preset accuracy rate is 95% in this preferred embodiment.

[0093] Specifically, the prediction accuracy of time series labels in the analysis results is the ratio of the number of accurately predicted time series labels to the total number of predictions in the analysis results.

[0094] In practice, the system of the present invention determines the compatibility of the fusion and the model by setting a preset accuracy rate, thereby reducing the impact of the decrease in the decision accuracy of the material feeding data due to the inaccuracy in determining the compatibility of the fusion and the model, and further improving the decision accuracy of the material feeding data.

[0095] Specifically, the increase in the overlap ratio of adjacent sliding windows is determined by the difference between the preset accuracy and the prediction accuracy of the time series labels in the analysis results.

[0096] Specifically, when the difference between the preset accuracy and the prediction accuracy of the time series labels in the analysis results is within 5%, the overlap ratio of adjacent sliding windows increases to 1.2 times the original value. When the difference between the preset accuracy and the prediction accuracy of the time series labels in the analysis results exceeds 5%, in addition to increasing to 1.2 times the original value, for every 1% exceeding 1%, the overlap ratio of adjacent sliding windows increases by 5%. For example, if the difference between the preset accuracy and the prediction accuracy of the time series labels in the analysis results is 7%, and the current overlap ratio of adjacent sliding windows is 30%, the increased overlap ratio of adjacent sliding windows will be 30 × 1.2 + 5 × 2 = 46%.

[0097] In implementation, the system of the present invention adjusts the overlap ratio of adjacent sliding windows by setting a preset accuracy rate. Since the fusion strategy adopts temporal sliding window fusion and the model uses a static CNN, the model cannot learn the temporal relationship within the window (such as the causal relationship of increasing impurities first and then decreasing flow rate). It can only treat the temporal data as static features, which leads to a disconnect between the temporal assumptions of the fusion strategy and the model's capabilities. By increasing the overlap ratio of adjacent sliding windows, the adjacent windows can share more time steps. In the prediction of continuous windows, the model can indirectly perceive the temporal correlation through the feature association of the overlapping parts, thereby alleviating the disconnect between the fusion strategy and the model's capabilities and further improving the decision accuracy of the feeding data.

[0098] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A multimodal feeding data analysis and decision-making system for an oil press based on AI optimization, characterized in that, include: The data acquisition module includes a data acquisition unit for acquiring multimodal feeding data of the oil press and a data transmission unit connected to the data acquisition unit for transmitting the feeding data to the processing position; The data processing module, which is connected to the data acquisition module, includes a preprocessing unit for preprocessing the feeding data to output feeding characteristics and a model training unit connected to the preprocessing unit for training an initial model based on the feeding characteristics to output a machine learning model. An analysis and decision-making module, which is connected to the data processing module, includes an analysis unit that analyzes the feeding data according to the machine learning model to output analysis results, and a decision-making unit connected to the analysis unit to make decisions based on the analysis results to optimize the feeding process of the oil press. A noise segmentation module, which is connected to the data processing module, is used to determine the adaptive noise band segmentation threshold based on the attenuation rate of residual noise in the feeding characteristics per unit time. A transmission buffer module, which is connected to the data acquisition module and the noise segmentation module respectively, is used to determine the adaptive buffer depth of the feeding data transmission fluctuation based on the data point matching failure rate of the feeding characteristics. A window adjustment module, which is connected to the data processing module and the transmission buffer module respectively, is used to determine the overlap ratio of adjacent sliding windows based on the prediction accuracy of the time series labels in the analysis results. The noise segmentation module responds to the attenuation rate of residual noise in the feeding characteristics being greater than or equal to a preset second attenuation rate, thus determining that the decision accuracy of the feeding data meets the requirements. The noise segmentation module responds to the fact that the attenuation rate of the residual noise in the feeding feature is less than the preset second attenuation rate, and determines that the decision accuracy of the feeding data does not meet the requirements. The noise segmentation module responds to the fact that the attenuation rate of the residual noise in the feeding feature is greater than the preset first attenuation rate and less than the preset second attenuation rate, and preliminarily determines that the synergy of multimodal fusion does not meet the requirements. The noise segmentation module adaptively adjusts the noise frequency band segmentation threshold in response to the attenuation rate of the residual noise in the feeding feature being less than or equal to the preset first attenuation rate. Specifically, when the intensity of high-frequency noise increases and the dominant frequency rises, the noise band segmentation threshold is increased; when the dominant noise frequency decreases, the noise band segmentation threshold is decreased.

2. The AI-optimized multimodal feeding data analysis and decision-making system for oil presses according to claim 1, characterized in that, The transmission buffer module responds to the data point matching failure rate of the feeding feature being less than or equal to a preset first failure rate, thus determining that the synergy of multimodal fusion meets the requirements; The transmission buffer module determines that the multimodal fusion coordination does not meet the requirements when the data point matching failure rate of the feeding feature is greater than the preset first failure rate.

3. The AI-optimized multimodal feeding data analysis and decision-making system for oil presses according to claim 2, characterized in that, The transmission buffer module increases the adaptive buffer depth for fluctuations in data point matching of the feeding feature when the failure rate is greater than the preset first failure rate and less than or equal to the preset second failure rate.

4. The AI-optimized multimodal feeding data analysis and decision-making system for oil presses according to claim 3, characterized in that, The transmission buffer module responds to the fact that the data point matching failure rate of the feeding feature is greater than the preset second failure rate, and preliminarily determines that the fusion and model adaptability do not meet the requirements.

5. The AI-optimized multimodal feeding data analysis and decision-making system for oil presses according to claim 4, characterized in that, The increase in the adaptive buffer depth for fluctuations in the feeding data transmission is determined by the difference between the failure rate of data point matching of the feeding features and a preset first failure rate.

6. The AI-optimized multimodal feeding data analysis and decision-making system for oil presses according to claim 5, characterized in that, The window adjustment module determines that the fusion and model fits the requirements if the prediction accuracy of the time series labels in the analysis results is greater than or equal to the preset accuracy. The window adjustment module responds to the fact that the prediction accuracy of the time series labels in the analysis results is less than the preset accuracy, determines that the fusion and model adaptability do not meet the requirements, and increases the overlap ratio of adjacent sliding windows.

7. The AI-optimized multimodal feeding data analysis and decision-making system for oil presses according to claim 6, characterized in that, The increase in the overlap ratio of adjacent sliding windows is determined by the difference between the preset accuracy and the prediction accuracy of the time series labels in the analysis results.

Images