A photovoltaic string communication anomaly identification method and system based on XGBoost

Abstract

This invention relates to a method and system for identifying communication anomalies in photovoltaic (PV) strings based on XGBoost. The method includes: acquiring the current values ​​of each PV string in a PV power plant during the test period and the current values ​​of all PV strings in the PV power plant during a preset period prior to the test period, along with corresponding communication anomaly tag data, and normalizing the current values; training an XGBoost model based on data generated by a VaDE model to obtain a trained XGBoost model; inputting the normalized current values ​​of each PV string in the PV power plant during the test period into the trained XGBoost model to identify PV strings in the PV power plant with abnormal current communication during the test period. The technical solution provided by this invention, by training an XGBoost model based on data generated by a VaDE model to obtain a trained XGBoost model, and using the trained model to identify whether the current communication of each string in a PV power plant is abnormal, can improve the accuracy of the XGBoost model in identifying abnormal PV strings in a PV power plant.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic string communication anomaly identification, and in particular to a photovoltaic string communication anomaly identification method and system based on XGBoost. Background Art

[0002] With the rapid development of science and technology, photovoltaic power generation technology has been widely used both at home and abroad. Its application forms are diverse and the application places are widely distributed. It is mainly used in large-scale ground photovoltaic power stations, roofs of residential and commercial buildings, building photovoltaic building integration, photovoltaic street lights, etc. In practical applications, due to the large number of photovoltaic strings in photovoltaic power stations, there is no way to timely discover and handle photovoltaic strings with communication abnormalities. Due to the communication abnormality, the data abnormality is sometimes present and sometimes absent, sometimes large and sometimes small, causing the string abnormality diagnosis model based on the data to have false detection and missed detection, reducing the credibility of the correlation analysis. Therefore, it is urgent to propose a method and system that can accurately identify photovoltaic string communication abnormalities in photovoltaic power stations. Summary of the invention

[0003] The present application provides a method and system for identifying photovoltaic string communication anomalies based on XGBoost, so as to at least solve the technical problem of inaccurate identification of whether photovoltaic string communication in a photovoltaic power station is abnormal in the related art.

[0004] The first embodiment of the present application proposes a photovoltaic string communication anomaly identification method based on XGBoost, the method comprising:

[0005] Obtain the current value of each photovoltaic string in the photovoltaic power station during the test period and the current value of all photovoltaic strings in the photovoltaic power station during a preset period before the test period and the corresponding communication abnormality tag data, and normalize the current values;

[0006] The normalized current values ​​of all photovoltaic strings in the photovoltaic power station within a preset period before the test date and the corresponding communication abnormality label data are recorded as sample set A, and the sample set consisting of the current values ​​of the faulty photovoltaic strings in A and the corresponding communication abnormality labels is recorded as B;

[0007] Input the current value of the faulty photovoltaic string in the sample set B into the pre-trained VaDE model to obtain a sample set of faulty photovoltaic string current value communication abnormalities generated by the VaDE model, and record the sample set as C, merge the sample set A and the sample set C to obtain a merged sample set D;

[0008] Use sample set D to train the XGBoost model to obtain a trained XGBoost model;

[0009] The normalized current value of each photovoltaic string in the photovoltaic power station during the test period is input into the trained XGBoost model to identify the photovoltaic strings with abnormal current value communication in the photovoltaic power station during the test period.

[0010] The second embodiment of the present application proposes a photovoltaic string communication anomaly identification system based on XGBoost, the system comprising:

[0011] An acquisition module is used to acquire the current value of each photovoltaic string in the photovoltaic power station during the test period and the current value of all photovoltaic strings in the photovoltaic power station during a preset period before the test period and the corresponding communication abnormality label data, and normalize the current value;

[0012] The first sample module is used to record the normalized current values ​​of all photovoltaic strings in the photovoltaic power station within a preset period before the test date and the corresponding communication abnormality label data as sample set A, and record the sample set consisting of the current values ​​of the faulty photovoltaic strings in A and the corresponding communication abnormality labels as B;

[0013] The second sample module is used to input the current value of the faulty photovoltaic string in the sample set B into the pre-trained VaDE model to obtain a sample set of faulty photovoltaic string current value communication abnormalities generated by the VaDE model, and record the sample set as C, merge the sample set A and the sample set C to obtain the merged sample set D;

[0014] The training module is used to train the XGBoost model using the sample set D to obtain a trained XGBoost model;

[0015] The identification module is used to input the normalized current value of each photovoltaic string in the photovoltaic power station during the test period into the trained XGBoost model to identify the photovoltaic strings with abnormal current value communication in the photovoltaic power station during the test period.

[0016] The technical solution provided by the embodiments of the present application brings at least the following beneficial effects:

[0017] The present invention provides a method and system for identifying photovoltaic string communication anomalies based on XGBoost, the method comprising: obtaining the current value of each photovoltaic string in a photovoltaic power station within a test period and the current values ​​of all photovoltaic strings in the photovoltaic power station within a preset period before the test period and the corresponding communication anomaly label data, and normalizing the current value; recording the normalized current values ​​of all photovoltaic strings in the photovoltaic power station within a preset period before the test date and the corresponding communication anomaly label data as a sample set A, and recording the sample set consisting of the current values ​​of the faulty photovoltaic strings in A and the corresponding communication anomaly labels as B; recording the current values ​​of the faulty photovoltaic strings in A and the corresponding communication anomaly labels as B; recording the current values ​​of the faulty photovoltaic strings in A and the corresponding communication anomaly labels as B. The current value of the faulty photovoltaic string in sample set B is input into the pre-trained VaDE model to obtain a sample set of abnormal current value communication of the faulty photovoltaic string generated by the VaDE model, and the sample set is recorded as C, and the sample set A and the sample set C are merged to obtain the merged sample set D; the XGBoost model is trained using sample set D to obtain a trained XGBoost model; the normalized value of the current of each photovoltaic string in the photovoltaic power station during the test period is input into the trained XGBoost model to identify the photovoltaic strings with abnormal current value communication in the photovoltaic power station during the test period. The technical solution provided by the present invention trains the XGBoost model based on the generated data of the VaDE model to obtain the trained XGBoost model, and uses the trained model to identify whether the current communication of each string in the photovoltaic power station is abnormal, which can improve the accuracy of the XGBoost model in identifying abnormal photovoltaic strings in the photovoltaic power station.

[0018] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0020] Figure 1 It is a flowchart of a photovoltaic string communication anomaly identification method based on XGBoost according to an embodiment of the present application;

[0021] Figure 2 It is a structural diagram of a photovoltaic string communication anomaly identification system based on XGBoost according to the present application. DETAILED DESCRIPTION

[0022] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0023] The present application proposes a method and system for identifying photovoltaic string communication anomalies based on XGBoost, the method comprising: obtaining the current value of each photovoltaic string in a photovoltaic power station within a test period and the current value of all photovoltaic strings in the photovoltaic power station within a preset period before the test period and the corresponding communication anomaly label data, and normalizing the current value; recording the normalized current value of all photovoltaic strings in the photovoltaic power station within a preset period before the test date and the corresponding communication anomaly label data as sample set A, and recording the sample set consisting of the current value of the faulty photovoltaic string in A and the corresponding communication anomaly label data as B; recording the current value of each photovoltaic string in the photovoltaic power station within a preset period before the test period and the corresponding communication anomaly label data as sample set A; recording the current value of each photovoltaic string in the photovoltaic power station within a preset period before the test period and the corresponding communication anomaly label data as sample set ... A; recording the current value of each photovoltaic string in the photovoltaic power station within a preset period before the test period and the corresponding communication anomaly label data as sample set B; recording the current value of each photovoltaic string in the photovoltaic power station within a preset period before the test period and the corresponding communication anomaly label data as sample set A; recording the current value of each photovoltaic string in the photovoltaic power The current value of the faulty photovoltaic string in sample set B is input into the pre-trained VaDE model to obtain a sample set of abnormal current value communication of the faulty photovoltaic string generated by the VaDE model, and the sample set is recorded as C, and the sample set A and the sample set C are merged to obtain the merged sample set D; the XGBoost model is trained using sample set D to obtain a trained XGBoost model; the normalized value of the current of each photovoltaic string in the photovoltaic power station during the test period is input into the trained XGBoost model to identify the photovoltaic strings with abnormal current value communication in the photovoltaic power station during the test period. The technical solution provided by the present invention trains the XGBoost model based on the generated data of the VaDE model to obtain the trained XGBoost model, and uses the trained model to identify whether the current communication of each string in the photovoltaic power station is abnormal, which can improve the accuracy of the XGBoost model in identifying abnormal photovoltaic strings in the photovoltaic power station.

[0024] Example 1

[0025] Figure 1 A flowchart of a photovoltaic string communication anomaly identification method based on XGBoost is provided in this embodiment, such as Figure 1 As shown, the method includes:

[0026] Step 1: obtaining the current value of each photovoltaic string in the photovoltaic power station during the test period and the current value of all photovoltaic strings in the photovoltaic power station during a preset period before the test period and the corresponding communication abnormality tag data, and normalizing the current value;

[0027] For example, the current value of each photovoltaic string in the photovoltaic power station on the detection date and the current values ​​of all photovoltaic strings in the photovoltaic power station within 30 days before the detection date and the corresponding communication abnormality label data are obtained, wherein the photovoltaic strings with current value communication abnormalities are included within the 30 days.

[0028] For example, the calculation formula for normalizing the current value of the photovoltaic string m at the jth moment in the test period is as follows:

[0029]

[0030] In the formula, is the normalized current value of the photovoltaic string m at the jth moment during the test period, x m,j is the current value of PV string m at the jth moment during the test period, x m,k is the current value of the photovoltaic string m at the kth moment in the test period, h is the set of all moments in the test period for obtaining data, and k is any moment in h.

[0031] Step 2: The normalized current values ​​of all photovoltaic strings in the photovoltaic power station within a preset period before the test date and the corresponding communication abnormality label data are recorded as sample set A, and the sample set consisting of the current values ​​of the faulty photovoltaic strings in A and the corresponding communication abnormality labels is recorded as B;

[0032] Step 3: Input the current value of the faulty photovoltaic string in the sample set B into the pre-trained VaDE model to obtain a sample set of faulty photovoltaic string current value communication abnormalities generated by the VaDE model, and record the sample set as C, merge the sample set A and the sample set C to obtain the merged sample set D;

[0033] It should be noted that the pre-trained VaDE model includes: a first neural network layer, a sampling layer and a second neural network layer;

[0034] The training process of the VaDE model includes:

[0035] Obtain normalized current value data of faulty photovoltaic strings in photovoltaic power stations within a historical period;

[0036] The acquired data is input into the first neural network layer, the sampling layer and the second neural network layer of the initial VaDE model, the marginal likelihood lower bound is used as the loss function of the model, and the model is trained using the adaptive matrix estimation Adam optimization algorithm to obtain a trained VaDE model.

[0037] Step 4: Use sample set D to train the XGBoost model to obtain a trained XGBoost model;

[0038] Step 5: Input the normalized current value of each photovoltaic string in the photovoltaic power station during the test period into the trained XGBoost model to identify the photovoltaic strings with abnormal current value communication in the photovoltaic power station during the test period.

[0039] In the embodiment of the present disclosure, the XGBoost model is trained using the sample set D to obtain a trained XGBoost model, including:

[0040] Get a sample set D, where D = {(x i ,y i)}(|D|=n,x i ∈R m ,y i ∈R), x i is the index value of the i-th sample at each time of the day, y i is the label value of whether there is communication anomaly in the indicator of the i-th sample day, y i ∈{0, 1}, 0 indicates normal communication, 1 indicates abnormal communication, n is the number of samples, and m is the number of sample features;

[0041] Based on the sample set D, a gradient boosting tree is constructed, and the CART regression tree is used as the subtree model of the model. The CART regression tree is added in an iterative manner, and all the CART regression trees are merged together to obtain the trained XGBoost model.

[0042] It should be noted that when adding a CART regression tree, a greedy algorithm is used to first find the segmentation point with the greatest benefit based on each sample feature, then find the feature with the greatest benefit based on all sample features, and pick out the segmentation point with the greatest gain in the double-layer loop of features and segmentation points, and split based on the segmentation point to add a CART regression tree.

[0043] Specifically, define the XGBoost model to predict y in D. First, construct a gradient boosting tree: accumulate trees based on the sample set D, build one tree each time, and use the CART regression tree as the subtree model of the model. Secondly, construct the loss function of the regularization term of the tth iteration: l is the true value y specified inside the algorithm i With the predicted value The loss function is a differentiable convex function, f t (x i ) is the weight of the i-th sample in the leaf node in the t-th iteration, Ω(f t ) is a regular term, which is represented by the number of leaf nodes and leaf node weights of the t-th iteration tree. Then, L (t) exist A second-order Taylor expansion is performed at , and the weights are determined by minimizing the loss function. Finally, all CART regression trees are merged together to obtain the trained XGBoost model.

[0044] It should be noted that the index value of the photovoltaic string obtained here is the current value, and other index values ​​of the photovoltaic string can also be obtained, which are not limited here.

[0045] In summary, the embodiment of the present disclosure provides a method for identifying photovoltaic string communication anomalies based on XGBoost. The XGBoost model is trained based on the generated data of the VaDE model to obtain a trained XGBoost model. The trained model is used to identify whether the current communication of each string in the photovoltaic power station is abnormal, which can improve the accuracy of the XGBoost model in identifying abnormal photovoltaic strings in the photovoltaic power station.

[0046] Example 2

[0047] Figure 2 A structural diagram of a photovoltaic string communication anomaly identification system based on XGBoost provided in an embodiment of the present disclosure, such as Figure 2 As shown, the system comprises:

[0048] An acquisition module is used to acquire the current value of each photovoltaic string in the photovoltaic power station during the test period and the current value of all photovoltaic strings in the photovoltaic power station during a preset period before the test period and the corresponding communication abnormality label data, and normalize the current value;

[0049] The first sample module is used to record the normalized current values ​​of all photovoltaic strings in the photovoltaic power station within a preset period before the test date and the corresponding communication abnormality label data as sample set A, and record the sample set consisting of the current values ​​of the faulty photovoltaic strings in A and the corresponding communication abnormality labels as B;

[0050] The second sample module is used to input the current value of the faulty photovoltaic string in the sample set B into the pre-trained VaDE model to obtain a sample set of faulty photovoltaic string current value communication abnormalities generated by the VaDE model, and record the sample set as C, merge the sample set A and the sample set C to obtain the merged sample set D;

[0051] The training module is used to train the XGBoost model using the sample set D to obtain a trained XGBoost model;

[0052] The identification module is used to input the normalized current value of each photovoltaic string in the photovoltaic power station during the test period into the trained XGBoost model to identify the photovoltaic strings with abnormal current value communication in the photovoltaic power station during the test period.

[0053] In the embodiment of the present disclosure, the calculation formula for normalizing the current value of the photovoltaic string m at the jth moment in the test period is as follows:

[0054]

[0055] In the formula, is the normalized current value of the photovoltaic string m at the jth moment during the test period, x m,jis the current value of PV string m at the jth moment during the test period, x m,k is the current value of the photovoltaic string m at the kth moment in the test period, h is the set of all moments in the test period for obtaining data, and k is any moment in h.

[0056] In the disclosed embodiment, the pre-trained VaDE model includes: a first neural network layer, a sampling layer, and a second neural network layer;

[0057] The training process of the VaDE model includes:

[0058] Obtain normalized current value data of faulty photovoltaic strings in photovoltaic power stations within a historical period;

[0059] The acquired data is input into the first neural network layer, the sampling layer and the second neural network layer of the initial VaDE model, the marginal likelihood lower bound is used as the loss function of the model, and the model is trained using the adaptive matrix estimation Adam optimization algorithm to obtain a trained VaDE model.

[0060] In the embodiment of the present disclosure, the training module includes:

[0061] The acquisition unit is used to acquire a sample set D, where D = {(x i ,y i )}(|D|=n,x i ∈R m ,y i ∈R), x i is the index value of the i-th sample at each time of the day, y i is the label value of whether there is communication anomaly in the indicator of the i-th sample day, y i ∈{0, 1}, 0 indicates normal communication, 1 indicates abnormal communication, n is the number of samples, and m is the number of sample features;

[0062] The training unit is used to construct a gradient boosting tree based on the sample set D, use the CART regression tree as the subtree model of the model, increase the CART regression tree in an iterative manner, and merge all the CART regression trees together to obtain the trained XGBoost model.

[0063] It should be noted that when adding a CART regression tree, a greedy algorithm is used to first find the segmentation point with the greatest benefit based on each sample feature, then find the feature with the greatest benefit based on all sample features, and pick out the segmentation point with the greatest gain in the double-layer loop of features and segmentation points, and split based on the segmentation point to add a CART regression tree.

[0064] In summary, the present invention provides a photovoltaic string communication anomaly identification system based on XGBoost, which trains the XGBoost model based on the generated data of the VaDE model to obtain a trained XGBoost model, and uses the trained model to identify whether the current communication of each string in the photovoltaic power station is abnormal, thereby improving the accuracy of the XGBoost model in identifying abnormal photovoltaic strings in the photovoltaic power station.

[0065] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0066] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0067] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A photovoltaic string communication anomaly identification method based on XGBoost, characterized in that: The method comprises: Obtain the current value of each photovoltaic string in the photovoltaic power station during the test period and the current value of all photovoltaic strings in the photovoltaic power station during a preset period before the test period and the corresponding communication abnormality tag data, and normalize the current values; The normalized current values ​​of all photovoltaic strings in the photovoltaic power station within a preset period before the test date and the corresponding communication abnormality label data are recorded as sample set A, and the sample set consisting of the current values ​​of the faulty photovoltaic strings in A and the corresponding communication abnormality labels is recorded as B; Input the current value of the faulty photovoltaic string in the sample set B into the pre-trained VaDE model to obtain a sample set of abnormal communication of the current value of the faulty photovoltaic string generated by the VaDE model, and record the sample set as C, merge the sample set A and the sample set C to obtain the merged sample set D, wherein: the pre-trained VaDE model includes: a first neural network layer, a sampling layer and a second neural network layer, and the process of training the VaDE model includes: obtaining normalized current value data of the faulty photovoltaic string in the photovoltaic power station within the historical period; inputting the obtained data into the first neural network layer, the sampling layer and the second neural network layer of the initial VaDE model, taking the lower bound of the marginal likelihood as the loss function of the model, and using the adaptive matrix estimation Adam optimization algorithm to train the model to obtain a trained VaDE model; Use sample set D to train the XGBoost model to obtain a trained XGBoost model; The normalized current value of each photovoltaic string in the photovoltaic power station during the test period is input into the trained XGBoost model to identify the photovoltaic strings with abnormal current value communication in the photovoltaic power station during the test period.

2. The method according to claim 1, characterized in that The calculation formula for normalizing the current value of the photovoltaic string m at the jth moment in the test period is as follows: In the formula, is the normalized current value of the photovoltaic string m at the jth moment during the test period, x m,j is the current value of PV string m at the jth moment during the test period, x m,k is the current value of the photovoltaic string m at the kth moment in the test period, h is the set of all moments in the test period for obtaining data, and k is any moment in h.

3. The method according to claim 1, characterized in that The XGBoost model is trained using the sample set D to obtain a trained XGBoost model, including: Get a sample set D, where D = {(x i ,y i )}(D|=n,x i ∈R m ,y i ∈R), x i is the index value of the i-th sample at each time of the day, y i is the label value of whether there is communication anomaly in the indicator of the i-th sample day, y i ∈{0, 1}, 0 indicates normal communication, 1 indicates abnormal communication, n is the number of samples, and m is the number of sample features; Based on the sample set D, a gradient boosting tree is constructed, and the CART regression tree is used as the subtree model of the model. The CART regression tree is added in an iterative manner, and all the CART regression trees are merged together to obtain the trained XGBoost model.

4. The method according to claim 3, characterized in that When adding a CART regression tree, a greedy algorithm is used to first find the segmentation point with the greatest benefit based on each sample feature, then find the feature with the greatest benefit based on all sample features, pick out the segmentation point with the greatest gain in the double-layer loop of features and segmentation points, and split based on the segmentation point to add a CART regression tree.

5. A photovoltaic string communication anomaly identification system based on XGBoost, characterized in that: The system comprises: An acquisition module is used to acquire the current value of each photovoltaic string in the photovoltaic power station during the test period and the current value of all photovoltaic strings in the photovoltaic power station during a preset period before the test period and the corresponding communication abnormality label data, and normalize the current value; The first sample module is used to record the normalized current values ​​of all photovoltaic strings in the photovoltaic power station within a preset period before the test date and the corresponding communication abnormality label data as sample set A, and record the sample set consisting of the current values ​​of the faulty photovoltaic strings in A and the corresponding communication abnormality labels as B; The second sample module is used to input the current value of the faulty photovoltaic string in the sample set B into the pre-trained VaDE model to obtain a sample set of abnormal communication of the current value of the faulty photovoltaic string generated by the VaDE model, and the sample set is recorded as C, and the sample set A and the sample set C are merged to obtain the merged sample set D, wherein: the pre-trained VaDE model includes: a first neural network layer, a sampling layer and a second neural network layer, and the process of training the VaDE model includes: obtaining normalized current value data of the faulty photovoltaic string in the photovoltaic power station within the historical period; inputting the obtained data into the first neural network layer, the sampling layer and the second neural network layer of the initial VaDE model, taking the lower bound of the marginal likelihood as the loss function of the model, and using the adaptive matrix estimation Adam optimization algorithm to train the model to obtain a trained VaDE model; The training module is used to train the XGBoost model using the sample set D to obtain a trained XGBoost model; The identification module is used to input the normalized current value of each photovoltaic string in the photovoltaic power station during the test period into the trained XGBoost model to identify the photovoltaic strings with abnormal current value communication in the photovoltaic power station during the test period.

6. The system according to claim 5, characterized in that The calculation formula for normalizing the current value of the photovoltaic string m at the jth moment in the test period is as follows: In the formula, is the normalized current value of the photovoltaic string m at the jth moment during the test period, x m,j is the current value of PV string m at the jth moment during the test period, x m,k is the current value of the photovoltaic string m at the kth moment in the test period, h is the set of all moments in the test period for obtaining data, and k is any moment in h.

7. The system according to claim 5, characterized in that The training module comprises: The acquisition unit is used to acquire a sample set D, where D = {(x i ,y i )}(D|=n,x i ∈R m ,y i ∈R), x i is the index value of the i-th sample at each time of the day, y i is the label value of whether there is communication anomaly in the indicator of the i-th sample day, y i ∈{0, 1}, 0 indicates normal communication, 1 indicates abnormal communication, n is the number of samples, and m is the number of sample features; The training unit is used to construct a gradient boosting tree based on the sample set D, use the CART regression tree as the subtree model of the model, increase the CART regression tree in an iterative manner, and merge all the CART regression trees together to obtain the trained XGBoost model.

8. The system according to claim 7, characterized in that When adding a CART regression tree, a greedy algorithm is used to first find the segmentation point with the greatest benefit based on each sample feature, then find the feature with the greatest benefit based on all sample features, pick out the segmentation point with the greatest gain in the double-layer loop of features and segmentation points, and split based on the segmentation point to add a CART regression tree.

Images