Energy-saving electro-hydraulic drive material distribution trolley and hydraulic integrated module control method

Abstract

The invention relates to the technical field of hydraulic integrated module control using models, in particular to an energy-saving type electro-hydraulic drive material distribution trolley and a hydraulic integrated module control method. The method comprises the following steps: firstly, analyzing the distribution centrality of electro-hydraulic driving parameters in a cluster based on state vectors to obtain an important coefficient; further analyzing the matching degree of samples in two clusters of each type of state vector in a limited state vector type cluster and an unlimited cluster based on electro-hydraulic driving parameters, and obtaining an important correction coefficient; further obtaining the influence degree according to the change trend of the initial training weight; and further correcting the initial training weight according to the training error of the current training in combination with the influence degree and the importance coefficient of the sample to obtain the corrected training weight of the next training until the model converges, so that more accurate and better trolley electro-hydraulic driving parameters can be provided in the prediction process, and the energy-saving effect is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of hydraulic integrated module control technology using models, specifically to an energy-saving electro-hydraulic driven material distribution trolley and a hydraulic integrated module control method. Background Technology

[0002] Electro-hydraulic drive enables the material handling trolley to operate stably in harsh environments with heavy loads, dust, and high vibration, with relatively simple maintenance and extended service life. The hydraulic module is the power core of the material handling trolley, which can adjust the flow rate and direction of hydraulic oil through the control valve group to achieve precise control of the direction of movement. Currently, machine learning is used to train the trolley with historical operating data and related parameters to obtain the optimal parameters for electro-hydraulic drive.

[0003] Because the historical data of the trolley is affected by its operating conditions, the trolley does not always operate according to its optimal parameters. Furthermore, the electro-hydraulic parameters may differ significantly between trolleys operating under similar conditions, making it difficult to determine the optimal parameters. The obtained parameters differ considerably from the trolley's intended optimal parameters, thus affecting the accuracy of electro-hydraulic drive control and leading to energy waste during operation, failing to achieve energy-saving goals. Summary of the Invention

[0004] To address the technical problem that historical data is affected by the trolley's operating status, leading to insufficient accuracy in machine learning-based control and impacting energy-saving performance, this invention aims to provide an energy-saving electro-hydraulic driven material distribution trolley and a hydraulic integrated module control method. The specific technical solution adopted is as follows:

[0005] A hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley, the method comprising:

[0006] Obtain historical material distribution data samples of the trolley, including state vectors and electro-hydraulic drive parameters;

[0007] In the clustering of samples based on the state vectors, the importance coefficient of each sample is obtained according to the distribution concentration of the electro-hydraulic drive parameters within each cluster; in the samples of each class of state vectors, clustering is performed based on the electro-hydraulic drive parameters to obtain a first class of clusters; all samples are clustered based on the electro-hydraulic drive parameters to obtain a second class of clusters; the matching degree of the samples in the first class of clusters and the second class of clusters corresponding to the samples of each class of state vectors is analyzed to obtain the importance correction coefficient of each sample; the importance coefficient and the importance correction coefficient are fused to obtain the initial training weights;

[0008] Based on the changing trend of the initial training weights corresponding to the same state vector of the same vehicle, the influence degree of each sample is obtained; based on the training error of the current training, combined with the influence degree of the sample and the importance coefficient, the initial training weights are corrected to obtain the corrected training weights of each sample in the next training process, until the model converges.

[0009] The current state vector of the vehicle is input into the trained model to obtain the electro-hydraulic drive parameters.

[0010] Furthermore, the method for obtaining the importance coefficient includes:

[0011] For each cluster: sort by the number of each type of electro-hydraulic driving parameter, and obtain the importance coefficient of the samples in each cluster based on the difference in the proportion of samples of the two types of electro-hydraulic driving parameters with the most and second most combined.

[0012] Furthermore, the method for obtaining the important correction coefficients includes:

[0013] In the samples of each class of state vectors: cluster matching is performed based on the proportion of common samples in the first and second class clusters; based on the similarity in quantity between the matched second class cluster and the first class cluster, and combined with the proportion of common samples at the time of matching, the important correction coefficient of each sample is obtained.

[0014] Furthermore, the method for obtaining the degree of influence includes:

[0015] The initial training weights of all samples of the same car are sorted in chronological order to construct a weight sequence. The initial training weights corresponding to the same state vector of each class are extracted and the order in the weight sequence is retained to construct a subsequence.

[0016] Extract the slope of the fitted line for each subsequence to obtain the influence factor of each subsequence; based on the overall characteristics of all influence factors of all cars, and combined with the influence factor of the subsequence to which each sample belongs, obtain the influence degree of each sample.

[0017] Furthermore, the method for obtaining the corrected training weights for each sample in the next training process includes:

[0018] During the initial training, the initial training weights are used as the corrected training weights; the electrohydraulic drive prediction parameters for each sample are obtained based on the corrected training weights currently being trained.

[0019] The prediction parameters are corrected based on the degree of influence and the importance coefficient, and the training error of each sample in the current training is obtained based on the difference between the corrected prediction parameters and the electro-hydraulic drive parameters of the sample.

[0020] By combining the training error and the initial training weights, the corrected training weights for each sample in the next training process are obtained.

[0021] Furthermore, the method for determining model convergence includes:

[0022] The mean squared error is used as the training error function. When the error reaches its minimum, the model is considered to have converged, and training of the model is stopped.

[0023] Furthermore, the method for obtaining the initial training weights includes:

[0024] The product of the importance coefficient of the sample and the importance correction coefficient is used as the initial training weight.

[0025] Furthermore, the method for obtaining the state vector includes:

[0026] The historical material distribution data samples include operating condition data, specifically including: material weight, trolley speed, hydraulic system pressure, vibration, and load current. The LeNet feature extraction network is used to obtain the state vector of the operating condition data for each sample.

[0027] Furthermore, the clustering is performed using the ISODATA algorithm.

[0028] The present invention also proposes an energy-saving electro-hydraulic driven material distribution trolley, including a trolley body containing a hydraulic integrated module, and a system including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of any one of the hydraulic integrated module control methods of the energy-saving electro-hydraulic driven material distribution trolley.

[0029] The present invention has the following beneficial effects:

[0030] This invention first acquires historical material distribution data samples of the vehicle, and then, based on this data, analyzes the distribution concentration of electro-hydraulic drive parameters in clusters based on state vectors to obtain importance coefficients, thus preliminarily assessing the focus of the samples and providing a basis for obtaining initial training weights. Furthermore, it analyzes the matching degree of samples within each cluster based on state vector categories, both with and without restrictions, to obtain important correction coefficients, preventing the model from learning only the average characteristics of most samples and ignoring special vehicles that still exhibit parameter differences under similar working conditions, thereby reducing prediction bias. Further, based on... By analyzing the changing trend of the initial training weights corresponding to the same state vector of the same vehicle, the impact of the vehicle's own performance changes is analyzed, and the degree of impact is obtained. This provides a basis for more accurate analysis of training errors and adjustment of training weights. Furthermore, based on the current training error, combined with the degree of impact and importance coefficient of the samples, the initial training weights are corrected to obtain the corrected training weights for the next training, until the model converges. This solves the problem that historical data is affected by the vehicle's operating state, leading to unsatisfactory machine learning results and inaccurate prediction and control. This enables the provision of more accurate and better electro-hydraulic drive parameters for the vehicle during the prediction process, ensuring energy-saving effects. Attached Figure Description

[0031] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 A flowchart illustrating a hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to an embodiment of the present invention;

[0033] Figure 2 A flowchart illustrating a method for obtaining the degree of influence according to an embodiment of the present invention;

[0034] Figure 3 This is a flowchart illustrating a method for obtaining corrected training weights in the next training process, as provided in one embodiment of the present invention. Detailed Implementation

[0035] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an energy-saving electro-hydraulic driven material distribution trolley and hydraulic integrated module control method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0037] The following description, in conjunction with the accompanying drawings, details the specific scheme of the energy-saving electro-hydraulic driven material distribution trolley and hydraulic integrated module control method provided by the present invention.

[0038] Please see Figure 1 The diagram illustrates a flowchart of a hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to an embodiment of the present invention, specifically including:

[0039] Step S1: Obtain historical material distribution data samples of the trolley, including state vectors and electro-hydraulic drive parameters.

[0040] In one embodiment of the present invention, the working condition data of the trolley in each material distribution is obtained through various sensors, including at least: material weight, trolley speed, hydraulic system pressure, vibration and load current; at the same time, the corresponding electro-hydraulic drive parameters are obtained, specifically the output current, and the data is time-series data.

[0041] A LeNet feature extraction network is used to obtain the state vector of the working condition data of each sample, thereby establishing the data sample of "state vector-electrohydraulic drive parameters" for each material distribution; a Transformer network is constructed, with the state vector as the input data and the corresponding electrohydraulic drive parameters as the output.

[0042] It should be noted that the historical data sample can be limited to the most recent year's historical data. The data collection frequency, collection method, and time range of historical data can all be set by the implementer according to actual needs. In other embodiments of the present invention, an autoencoder can also be used to construct an encoder-decoder structure. The original data is compressed through unsupervised learning. The output of the encoder's hidden layer is the state vector. It, along with the LeNet feature extraction network and the Transformer network, are well-known technologies and will not be described in detail here.

[0043] Step S2: In the clusters where samples are clustered based on state vectors, obtain the importance coefficient of each sample according to the distribution concentration of electro-hydraulic drive parameters within each cluster; in the samples of each class of state vectors, cluster based on electro-hydraulic drive parameters to obtain the first class of clusters; cluster all samples based on electro-hydraulic drive parameters to obtain the second class of clusters; analyze the matching degree of the samples in the first and second classes of clusters corresponding to the samples of each class of state vectors to obtain the importance correction coefficient of each sample; fuse the importance coefficient and the importance correction coefficient to obtain the initial training weights.

[0044] In clustering based on state vectors, vehicle data samples with similar working states are grouped into one class. Since changes in electro-hydraulic drive parameters directly affect the working state of the vehicle, the electro-hydraulic drive parameters of data samples with similar working states should also be concentrated. Therefore, the concentration of electro-hydraulic drive parameters within each cluster reflects the normal probability of the electro-hydraulic drive parameters, thereby obtaining the importance coefficient of each sample, initially assessing the degree of focus of the samples, and providing a basis for obtaining the initial training weights.

[0045] Preferably, in one embodiment of the present invention, in a cluster that clusters samples based on state vectors, for each cluster: firstly, each type of electro-hydraulic driving parameter can be regarded as a sub-cluster, and sorted by the number of corresponding electro-hydraulic driving parameters of each type to obtain the sorting sequence of all sub-clusters in a cluster;

[0046] Considering that the proportion of samples of the most numerous electro-hydraulic driving parameters is higher, while the proportion of samples of the second most numerous electro-hydraulic driving parameters is lower, it indicates that the electro-hydraulic driving parameters are concentrated in a sub-cluster. The stronger the concentration of distribution, the higher the accuracy of judging electro-hydraulic driving parameters under similar state vectors, and thus the smaller the sample specificity, and the smaller its importance in subsequent analysis.

[0047] Based on this, the importance coefficient of samples within each cluster is obtained by considering the difference in the proportion of samples with the most and second most combined electro-hydraulic drive parameters.

[0048] In clusters that cluster samples based on state vectors, for each cluster: the proportion of the number of samples of a certain type of electro-hydraulic driving parameter to the total number of samples in the entire cluster is taken as the quantity ratio; the difference between the quantity ratio of the most numerous electro-hydraulic driving parameter and the quantity ratio of the second most numerous electro-hydraulic driving parameter is taken as the numerator, and the quantity ratio of the second most numerous electro-hydraulic driving parameter is taken as the denominator. After linearly normalizing the fractional ratio as the independent variable in the corresponding data dimension, the difference between the constant 1 and the normalization result is taken as the importance coefficient of each sample in each cluster.

[0049] Clustering is performed using the ISODATA algorithm, with identical electrohydraulic drive parameters grouped into one class. The distribution of the number of samples of each class of electrohydraulic drive parameters within a cluster is represented by the proportion of samples. By measuring the relative deviation, the difference in the proportion of the two classes of electrohydraulic drive parameters with the largest and second largest number of samples is shown, thus reflecting the concentration of the distribution of electrohydraulic drive parameters within the cluster. The larger the normalization result, the stronger the concentration of the distribution. Then, a negative correlation mapping is performed using a constant of 1 to obtain the importance coefficient of each sample within the cluster. The analysis method for each cluster based on the state vector clustering of samples is consistent, and only one example is described here without repeating the explanation.

[0050] In other embodiments of the present invention, the implementer may also perform negative correlation normalization mapping using the exp(-x) function with the natural constant e as the base, taking the ratio of the fraction to be mapped as the independent variable of the function and the result of the function mapping as the importance coefficient; x is the independent variable, and the ISODATA algorithm is a well-known technique, which will not be described in detail here.

[0051] However, due to inherent factors of the trolley itself, such as usage time and transport load, the trolley's operating state under similar conditions may exhibit significantly different electro-hydraulic drive parameters. This indicates that even minor differences in the trolley's state characteristics can lead to substantial variations in the electro-hydraulic drive parameters, resulting in inaccurate predictions of the material handling trolley's electro-hydraulic drive parameters. Therefore, it is necessary to further analyze the differences in electro-hydraulic drive parameters under similar state vectors to further assess the importance of the samples.

[0052] Considering that clustering based on electro-hydraulic drive parameters is performed on the samples of each state vector to obtain the first cluster, which can reflect the internal difference distribution of parameters under the limited working conditions; at the same time, clustering all samples based on electro-hydraulic drive parameters to obtain the second cluster reflects the overall distribution pattern of parameters under different working conditions; then, analyzing the matching degree of the samples in the first and second clusters corresponding to the samples of each state vector can reveal whether there are differences in drive parameters between different vehicles under the same or similar working conditions. Such differences often come from individual factors such as the service life of the vehicle, load conditions, and wear degree. Through this local-global comparison, the important correction coefficients of each sample are obtained, avoiding the model only learning the average characteristics of most samples and ignoring special vehicles that still show parameter differences under similar working conditions, reducing prediction bias and improving prediction accuracy.

[0053] Preferably, in one embodiment of the present invention, the same state vector is a class; in the samples of each class of state vectors: considering that the greater the proportion of common samples between different classes of clusters, the higher the matching degree, cluster matching is first performed according to the proportion of common samples in the first class of clusters and the second class of clusters;

[0054] As an example, the cross-union ratio (CUI) of samples within two clusters represents the proportion of common samples in the first and second clusters, characterizing the matching degree of samples within the two clusters. The first cluster is matched with the second cluster corresponding to its maximum CUI. The CUI of each first cluster and each second cluster is calculated one by one to obtain the matching second cluster for each first cluster. Different first clusters can be matched with the same second cluster at the same time.

[0055] The intersection-union ratio is a fractional ratio obtained by taking the intersection of samples within two clusters as the numerator and the union as the denominator.

[0056] Considering that the more second-class clusters matched, the closer they are to the number of first-class clusters, it indicates that there are significant differences in electro-hydraulic drive parameters under the same state vector. The corresponding samples need more attention, and the higher the importance correction coefficient is. At the same time, the higher the proportion of common samples during matching, the higher the degree of matching and the higher the credibility of the matching relationship. Based on this, the importance correction coefficient of each sample is obtained according to the similarity between the number of second-class clusters and first-class clusters matched, combined with the proportion of common samples during matching.

[0057] In the samples of each state vector: the number of the second type of clusters to be matched is used as the numerator, the number of the first type of clusters is used as the denominator, and the fractional ratio is used as the quantitative similarity parameter, representing the quantitative similarity between the matched second type of clusters and the first type of clusters; the product of the proportion of common samples (maximum crossover ratio) when each first type of cluster is matched and the quantitative similarity parameter is used as the important correction coefficient for each sample in each first type of cluster.

[0058] The analysis method for samples of each type of state vector is the same. Only one example will be described here, and it will not be repeated.

[0059] It should be noted that before clustering the electro-hydraulic drive parameters, the DTW distance of the electro-hydraulic drive parameters of different samples is first calculated to obtain the DTW distance matrix, and then the ISODATA algorithm is used for clustering. In other embodiments of the present invention, hierarchical clustering can also be used to cluster the state vectors, DTW distance can be used to replace Euclidean distance, and K-Means can be combined to cluster the electro-hydraulic drive parameters. These are all existing technologies and will not be described in detail here.

[0060] The importance coefficient reflects the particularity of a sample in the overall distribution among similar state vectors and can reflect the importance of state features; the importance correction coefficient is used to correct the driving parameter offset caused by individual differences of the vehicle and reflects the sensitivity and reference value of the sample for model training under similar working conditions. Therefore, the importance coefficient and the importance correction coefficient are fused. At the same time, in order to accelerate the training process of the neural network, the fused result is used as the initial training weight of the sample.

[0061] Preferably, in one embodiment of the present invention, the product of the importance coefficient and the importance correction coefficient of the sample is used as the initial training weight.

[0062] In other embodiments of the present invention, the average value or weighted sum of the importance coefficients and importance correction coefficients of the samples can also be used as the initial training weights.

[0063] Step S3: Based on the changing trend of the initial training weights corresponding to the same state vector of the same car, obtain the influence degree of each sample; based on the current training error, combined with the influence degree and importance coefficient of the sample, correct the initial training weights to obtain the corrected training weights of each sample in the next training process, until the model converges.

[0064] Since different vehicles have been used for different periods of time—for example, some vehicles may have just been put into use while others have already reached their service life—and the same state vector of the same vehicle represents the same working condition of the vehicle, when the vehicle is used multiple times, its performance may decrease, and the corresponding electro-hydraulic drive parameters may also change. As a result, the initial training weights will also change accordingly. Therefore, by measuring the changing trend of the initial training weights corresponding to the same state vector of the same vehicle, we can obtain the degree of influence of each sample, which provides a basis for more accurate analysis of training errors and adjustment of training weights.

[0065] Preferably, in one embodiment of the present invention, please refer to Figure 2 The flowchart illustrates a method for obtaining the degree of influence provided by an embodiment of the present invention, specifically including:

[0066] Step S301: Sort the initial training weights of all samples of the same car in chronological order to construct a weight sequence, extract the initial training weights corresponding to the same state vector of each class and retain the order in the weight sequence to construct a subsequence.

[0067] In one embodiment of the present invention, in order to analyze the changes in the initial training weights, the initial training weights of all samples of the same car in chronological order are mapped to a two-dimensional coordinate system with the time axis as the horizontal axis, the data axis of the initial training weights as the vertical axis, the termination time point of each sample as the horizontal axis coordinate, and the initial training weights as the vertical axis coordinate, and the initial training weights are mapped to the vertical axis coordinate to obtain the weight sequence.

[0068] Choose any one type of state vector, remove the data points of other state vectors from the coordinate system, and keep only the data points of the selected state vector to obtain a subsequence.

[0069] The analysis method for each state vector of each car and the analysis method for the degree of influence on each sample are the same. Only one example is described here, and it will not be repeated.

[0070] Step S302: Extract the slope of the fitted line of each subsequence to obtain the influence factor of each subsequence; based on the overall characteristics of all influence factors of all cars, combined with the influence factor of the subsequence to which each sample belongs, obtain the influence degree of each sample.

[0071] Considering that the performance of the trolley will decrease after multiple uses, larger electro-hydraulic drive parameters are needed to meet the material distribution or transportation under the same conditions. At this time, the distribution of electro-hydraulic drive parameters is less concentrated. At the same time, when matching clusters in important correction coefficients, more second-class clusters are matched, which leads to a gradual increase in the initial training weights. Therefore, the slope of the fitted line of each subsequence is extracted to obtain the influence factor of each subsequence.

[0072] Considering that the overall characteristics of the influence factors of all subsequences of all cars represent the general performance degradation pattern across the entire fleet under different service years and operating conditions, and represent the overall weight change pattern, we can obtain the influence degree of each sample by combining the overall characteristics of the influence factors of all cars.

[0073] As an example, the slope of the fitted line is extracted by fitting a straight line using the least squares method. To prevent the slope of the fitted line from being negative, the slope is mapped as an independent variable using the exp(x) function, and the mapping result is used as the influence factor of the subsequence. The average value is used to represent the overall characteristics of the influence factor. In order to ensure that the degree of influence can correct the training error in the future, the sum of the average value of all influence factors and the influence factor of the subsequence to which each sample belongs, plus a constant of 1, is used as the influence degree of each sample, and the influence degree is a specific data value.

[0074] It should be noted that when the length of the subsequence is 1 and the slope cannot be extracted, the influence level of the corresponding sample is set to 1. The overall characteristics of the influence factors can also be represented by weighted fusion of the mean, median and mode of all influence factors. The least squares method, the method of obtaining the slope and the exponential function exp(x) with the natural constant e as the base are all well-known techniques and will not be elaborated further.

[0075] Furthermore, considering that if the vehicle distributes materials multiple times, the majority of samples will have their performance affected, and that the first sample in the time series of the vehicle is not always the first time the vehicle is put into use, the result of "clustering based on electro-hydraulic drive parameters" will have a certain bias, resulting in an unclear trend in the change of the initial training weights. At the same time, since the importance coefficient represents the importance of the sample, the initial training weights are also corrected by combining the current training error and the importance coefficient, and the corrected training weights for each sample in the next training process are obtained together until the model converges.

[0076] Preferably, in one embodiment of the present invention, please refer to Figure 3 The flowchart illustrates a method for obtaining corrected training weights in the next training process according to an embodiment of the present invention, specifically including:

[0077] Step S311: During the first training, the initial training weights are used as the corrected training weights; based on the corrected training weights of the current training, the electro-hydraulic drive prediction parameters for each sample are obtained.

[0078] To facilitate the description and construction of the loop, the initial training weights are used as the corrected training weights during the first training. This way, there are corresponding corrected training weights for each training session. Based on the corrected training weights of the current training, the state vector of each sample is input, and the electro-hydraulic driving prediction parameters of each sample are obtained using the Transformer network.

[0079] Step S312: Correct the prediction parameters according to the degree of influence and importance coefficient, and obtain the training error of each sample in the current training based on the difference between the corrected prediction parameters and the electro-hydraulic drive parameters of the sample.

[0080] Considering that a greater degree of influence indicates a greater impact of vehicle performance degradation on electro-hydraulic drive parameters, leading to a smaller predicted electro-hydraulic drive parameters, and a larger sample importance coefficient, it indicates a poorer distribution concentration of electro-hydraulic drive parameters, a greater impact of performance on the sample's electro-hydraulic drive parameters, and a larger correction magnitude for the predicted parameters, the predicted parameters are corrected based on the degree of influence and importance coefficient, thereby obtaining the training error of each currently trained sample based on the corrected predicted parameters.

[0081] As an example, for each sample, the product of the influence degree and importance coefficient is linearly normalized in the corresponding data dimension. The sum of the normalization result and the constant 1 is used as a correction factor. The product of the correction factor and the prediction parameter is used as the corrected prediction parameter. The absolute value of the difference between the corrected prediction parameter and the sample's electrohydraulic drive parameter is used as the numerator, and the sample's electrohydraulic drive parameter is used as the denominator. After the fractional ratio is linearly normalized in the corresponding data dimension, the normalization result is used as the training error for each sample in the current training.

[0082] The influence degree and importance coefficient are fused by multiplication, and the correction factor is guaranteed to be greater than 1 by using a constant of 1. The absolute difference between the corrected predicted parameters and the electro-hydraulic drive parameters of the samples is expressed by the absolute value of the difference, and the relative difference is expressed by the fractional ratio, thereby obtaining the training error.

[0083] It should be noted that, in one embodiment of the present invention, the data dimension corresponding to linear normalization is the data dimension composed of the corresponding parameters of all samples.

[0084] Step S313: Combine the training error and the initial training weights to obtain the corrected training weights for each sample in the next training process.

[0085] The larger the training error, the more attention should be paid to this sample in the next training, thus increasing the training weight of this sample in the next training process. Therefore, the training error and the initial training weight are combined in this way.

[0086] As an example, since the training error has been normalized, the larger the training error, the more attention should be paid in the next training, and the larger the corrected training weight should be. In order to ensure the effect of training error correction, the sum of training error and constant 1 is used as the correction coefficient. The product of the correction coefficient of each sample and the initial training weight is used as the corrected training weight of each sample in the next training process.

[0087] Preferably, in one embodiment of the present invention, the mean squared error is used as the training error function. When the error reaches its minimum, the model is determined to have converged, and the training of the model is stopped, thus obtaining a trained neural network model.

[0088] Step S4: Input the current state vector of the vehicle into the trained model to obtain the electro-hydraulic drive parameters.

[0089] After training the model through steps S1-S3, the current state vector of the trolley can be input into the trained model to obtain the electro-hydraulic drive parameters, thereby realizing the control of the material dispensing trolley.

[0090] Specifically, the vehicle's operating data is converted into state vectors and input into the trained model, which outputs electro-hydraulic drive parameters, providing more accurate and better electro-hydraulic drive parameters for the vehicle and ensuring energy-saving performance.

[0091] An embodiment of the present invention also provides an energy-saving electro-hydraulic driven material distribution trolley, including a trolley body containing a hydraulic integrated module, and further including a memory, a processor, and a computer program. The memory is used to store the corresponding computer program, and the processor is used to run the corresponding computer program. When the computer program runs in the processor, it can realize the hydraulic integrated module control method of the energy-saving electro-hydraulic driven material distribution trolley described in steps S1-S4.

[0092] In summary, to address the technical problem that historical data is affected by the vehicle's operating state, leading to insufficient accuracy in machine learning control and impacting energy-saving performance, this invention provides an energy-saving electro-hydraulic driven material handling vehicle and a hydraulic integrated module control method. This invention first acquires historical material handling data samples from the vehicle; then, it analyzes the distribution concentration of electro-hydraulic drive parameters within clusters based on state vector clustering to obtain importance coefficients; further, it analyzes the matching degree of samples within each cluster based on state vector categories and unrestricted clusters based on electro-hydraulic drive parameters to obtain importance correction coefficients; further, it obtains the degree of influence based on the changing trend of the initial training weights; and finally, based on the current training error, combined with the degree of influence and importance coefficients of the samples, it corrects the initial training weights to obtain corrected training weights for the next training iteration until the model converges. This solves the problem that historical data is affected by the vehicle's operating state, leading to unsatisfactory machine learning results and inaccurate prediction and control. This allows for more accurate and optimized electro-hydraulic drive parameters during prediction, ensuring energy-saving performance.

[0093] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0094] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley, characterized in that, The method includes: Obtain historical material distribution data samples of the trolley, including state vectors and electro-hydraulic drive parameters; In the clustering of samples based on the state vectors, the importance coefficient of each sample is obtained according to the distribution concentration of the electro-hydraulic drive parameters within each cluster; in the samples of each class of state vectors, clustering is performed based on the electro-hydraulic drive parameters to obtain a first class of clusters; all samples are clustered based on the electro-hydraulic drive parameters to obtain a second class of clusters; the matching degree of the samples in the first class of clusters and the second class of clusters corresponding to the samples of each class of state vectors is analyzed to obtain the importance correction coefficient of each sample; the importance coefficient and the importance correction coefficient are fused to obtain the initial training weights; Based on the changing trend of the initial training weights corresponding to the same state vector of the same vehicle, the influence degree of each sample is obtained; based on the training error of the current training, combined with the influence degree of the sample and the importance coefficient, the initial training weights are corrected to obtain the corrected training weights of each sample in the next training process, until the model converges. The current state vector of the vehicle is input into the trained model to obtain the electro-hydraulic drive parameters.

2. The hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to claim 1, characterized in that, The methods for obtaining the importance coefficients include: For each cluster: sort by the number of each type of electro-hydraulic driving parameter, and obtain the importance coefficient of the samples in each cluster based on the difference in the proportion of samples of the two types of electro-hydraulic driving parameters with the most and second most combined.

3. The hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to claim 1, characterized in that, The methods for obtaining the important correction coefficients include: In the samples of each class of state vectors: cluster matching is performed based on the proportion of common samples in the first and second class clusters; based on the similarity in quantity between the matched second class cluster and the first class cluster, and combined with the proportion of common samples at the time of matching, the important correction coefficient of each sample is obtained.

4. The hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to claim 1, characterized in that, The methods for obtaining the degree of influence include: The initial training weights of all samples of the same car are sorted in chronological order to construct a weight sequence. The initial training weights corresponding to the same state vector of each class are extracted and the order in the weight sequence is retained to construct a subsequence. Extract the slope of the fitted line for each subsequence to obtain the influence factor of each subsequence; based on the overall characteristics of all influence factors of all cars, and combined with the influence factor of the subsequence to which each sample belongs, obtain the influence degree of each sample.

5. The hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to claim 1, characterized in that, The method for obtaining the corrected training weights for each sample in the next training process includes: During the initial training, the initial training weights are used as the corrected training weights; the electrohydraulic drive prediction parameters for each sample are obtained based on the corrected training weights currently being trained. The prediction parameters are corrected based on the degree of influence and the importance coefficient, and the training error of each sample in the current training is obtained based on the difference between the corrected prediction parameters and the electro-hydraulic drive parameters of the sample. By combining the training error and the initial training weights, the corrected training weights for each sample in the next training process are obtained.

6. The hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to claim 1, characterized in that, The methods for determining model convergence include: The mean squared error is used as the training error function. When the error reaches its minimum, the model is considered to have converged, and training of the model is stopped.

7. The hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to claim 1, characterized in that, The method for obtaining the initial training weights includes: The product of the importance coefficient of the sample and the importance correction coefficient is used as the initial training weight.

8. The hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to claim 1, characterized in that, The method for obtaining the state vector includes: The historical material distribution data samples include operating condition data, specifically including: material weight, trolley speed, hydraulic system pressure, vibration, and load current. The LeNet feature extraction network is used to obtain the state vector of the operating condition data for each sample.

9. The hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley according to claim 1, characterized in that, The clustering was performed using the ISODATA algorithm.

10. An energy-saving electro-hydraulic driven material dispensing trolley, comprising a trolley body containing a hydraulic integrated module, and further comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the hydraulic integrated module control method for an energy-saving electro-hydraulic driven material distribution trolley as described in any one of claims 1 to 9.

Images