New energy station power prediction method and system based on longitudinal federal decision tree
By combining dynamic segmentation strategy and nonlinear spline transformation in the vertical federated decision tree model, a lightweight order-preserving encryption scheme is constructed, which solves the balance between security and efficiency in vertical federated learning and achieves efficient privacy protection and accurate power prediction of new energy power plants.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG UNIV
- Filing Date
- 2026-03-09
- Publication Date
- 2026-05-29
AI Technical Summary
Existing privacy computing technologies struggle to achieve a good balance between security and efficiency in vertical federated learning. Traditional order-preserving encryption schemes pose a risk of information leakage, and complex cryptographic operations can impair model accuracy or increase computational burden.
By combining dynamic segmentation strategy, logical interval random permutation and nonlinear monotonic spline transformation within segments, a lightweight order-preserving encryption scheme is constructed through a vertical federated decision tree model. This scheme maintains the topological structure of feature value sorting while obfuscating the statistical distribution and specific values of the original features.
It achieves efficient privacy protection in vertical federated learning, resists distribution alignment attacks, maintains model accuracy, reduces computational and communication overhead, and is suitable for large-scale high-dimensional data scenarios.
Smart Images
Figure CN121786465B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the fields of federated learning and information security technology, specifically to a method and system for predicting the power of new energy power plants based on a vertical federated decision tree. Background Technology
[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.
[0003] In the wave of digital transformation, data has become a core production factor. In scenarios such as financial risk control, joint marketing, and medical consultations, it is often necessary to aggregate feature data from different institutions for joint modeling. Because the participating parties possess different and complementary feature dimensions, this scenario is known as Vertical Federated Learning (VFL). However, due to the limitation that participating parties cannot directly exchange plaintext raw data, privacy leakage risks have become a bottleneck restricting the collaborative value creation of multi-party data.
[0004] Among existing privacy-preserving computation technologies, homomorphic encryption (HE) suffers from extremely low training efficiency due to excessive computational overhead; multi-party computation (MPC) faces significant latency challenges in network communication. While order-preserving encryption (OPE) can maintain the total order of data to facilitate direct training of decision trees (such as XGBoost and LightGBM), traditional order-preserving encryption schemes are vulnerable to privacy attacks such as statistical alignment. Attackers can observe the probability density distribution of the encrypted data and, combined with prior background knowledge, reconstruct the physical meaning of the original features.
[0005] Among numerous machine learning models, tree models (such as decision trees, random forests, and gradient boosting trees) continue to play an irreplaceable role in key areas such as financial scoring, risk assessment, and recommender systems due to their high interpretability, ease of parallelization, and insensitivity to numerical scaling. The construction of tree models essentially depends only on the relative order of feature values, and is independent of their absolute size. This characteristic provides an important opportunity to design lighter and more secure privacy-preserving methods. However, existing technical solutions still have the following limitations:
[0006] (1) Existing solutions attempt to protect the sequence of features using order-preserving transformations, but most of them fail to achieve a good balance between security and efficiency;
[0007] (2) The statistical distribution and numerical patterns of the original characteristics have not been completely broken, and there is still a risk of information leakage;
[0008] (3) Introducing too much noise or complex cryptographic operations can seriously impair the accuracy of the model or significantly increase the computational burden. Summary of the Invention
[0009] To address the aforementioned issues, this disclosure proposes a power prediction method and system for new energy power plants based on a vertical federated decision tree. By combining dynamic segmentation strategy, logical interval random permutation, and nonlinear monotonic spline transformation within segments, the method completely obfuscates the statistical distribution and specific values of the original features while fully maintaining the topological structure of the feature value sorting, thereby improving the efficiency of joint power prediction and enabling more accurate scheduling optimization of new energy power plants.
[0010] According to some embodiments, the present disclosure adopts the following technical solutions:
[0011] A power prediction method for renewable energy power plants based on vertical federated decision trees includes:
[0012] Acquire real-time environmental feature data;
[0013] Environmental characteristic data is input into the power load prediction model, and the predicted power value is output.
[0014] The power load forecasting model is a vertical federated decision tree model. The training process of the vertical federated decision tree model is as follows: each participant generates a key and hyperparameters for each column of features according to security requirements, and generates a random topology mapping for each column of features. A nonlinear transformation function is constructed based on each mapping relationship. The generated nonlinear transformation function and lightweight order-preserving encryption are used to cascade encrypted data. The encrypted data is concentrated in the coordinator for encrypted state model training. After training is completed, the vertical federated decision tree model is obtained.
[0015] According to some embodiments, the present disclosure adopts the following technical solutions:
[0016] A power prediction system for renewable energy power plants based on a vertical federated decision tree includes:
[0017] The data acquisition module is used to acquire real-time environmental characteristic data;
[0018] The load forecasting module is used to input environmental characteristic data into the power load forecasting model and output the predicted power value.
[0019] The power load forecasting model is a vertical federated decision tree model. The training process of the vertical federated decision tree model is as follows: each participant generates a key and hyperparameters for each column of features according to security requirements, and generates a random topology mapping for each column of features. A nonlinear transformation function is constructed based on each mapping relationship. The generated nonlinear transformation function and lightweight order-preserving encryption are used to cascade encrypted data. The encrypted data is concentrated in the coordinator for encrypted state model training. After training is completed, the vertical federated decision tree model is obtained.
[0020] According to some embodiments, the present disclosure adopts the following technical solutions:
[0021] A computer program product includes a computer program that, when executed by a processor, implements the aforementioned power prediction method for new energy power plants based on a vertical federated decision tree.
[0022] According to some embodiments, the present disclosure adopts the following technical solutions:
[0023] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned power prediction method for new energy power plants based on a vertical federated decision tree.
[0024] According to some embodiments, the present disclosure adopts the following technical solutions:
[0025] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the power prediction method for new energy power plants based on a vertical federated decision tree.
[0026] Compared with the prior art, the beneficial effects of this disclosure are as follows:
[0027] This disclosure presents a power prediction method for renewable energy power plants based on a vertical federated decision tree. This method combines a dynamic segmentation strategy, logical interval random permutation, and intra-segment nonlinear monotonic spline transformation to completely obfuscate the statistical distribution and specific values of the original features while fully preserving the topological structure of the feature value ranking. This disclosure aims to avoid the use of computationally expensive differential privacy noise or complex cryptographic protocols, achieving privacy-preserving multi-party modeling with near-plaintext training accuracy.
[0028] This disclosed method for predicting the power of renewable energy power plants based on vertical federated decision trees combines piecewise permutation and nonlinear splines for dual protection, effectively masking the probability density function of features and resisting distribution alignment attacks. It also provides a privacy-preserving tree model training scheme that simultaneously achieves the following objectives in vertical federated learning: first, completely hiding the distribution and specific values of the original features; second, strictly maintaining the relative order of feature values to support efficient tree splitting; and third, introducing only extremely low computational and communication overhead, making it suitable for large-scale, high-dimensional data scenarios.
[0029] This disclosed method for predicting the power of renewable energy power plants based on a vertical federated decision tree maintains global order mapping. The splitting decisions of the decision tree model on encrypted text are completely equivalent to those on plaintext, resulting in almost zero model accuracy loss. The entire transformation process involves only polynomial operations and a one-time preprocessing step, eliminating the need for expensive encrypted computations during training. This method is highly practical for large-scale datasets. Each participant has complete control over its local mapping parameters and permutation vectors; the central party cannot recover any original values through decryption, ensuring autonomy and control. Attached Figure Description
[0030] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.
[0031] Figure 1 This is a flowchart of the power prediction method for new energy power plants based on a vertical federated decision tree, according to an embodiment of this disclosure.
[0032] Figure 2 The system structure for power prediction of new energy power plants in the power system is shown in this embodiment of the present disclosure. Detailed Implementation
[0033] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.
[0034] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, 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 disclosure pertains.
[0035] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0036] Example 1
[0037] One embodiment of this disclosure provides a power prediction method for new energy power plants based on a vertical federated decision tree. The method includes the following steps:
[0038] Step 1: Obtain real-time environmental feature data;
[0039] Step 2: Input environmental characteristic data into the power load prediction model and output the predicted power value;
[0040] The power load forecasting model is a vertical federated decision tree model. The training process of the vertical federated decision tree model is as follows: each participant generates a key and hyperparameters for each column of features according to security requirements, and generates a random topology mapping for each column of features. A nonlinear transformation function is constructed based on each mapping relationship. The generated nonlinear transformation function and lightweight order-preserving encryption are used to cascade encrypted data. The encrypted data is concentrated in the coordinator for encrypted state model training. After training is completed, the vertical federated decision tree model is obtained.
[0041] As one embodiment, the power prediction method for new energy power plants based on vertical federated decision trees disclosed herein, through a combination of dynamic segmentation strategy, logical interval random permutation, and nonlinear monotonic spline transformation within segments, completely obfuscates the statistical distribution and specific values of the original features while fully preserving the topological structure of the feature value sorting. The specific implementation process is as follows:
[0042] Step 1: Obtain real-time environmental feature data;
[0043] The power plants possess local real-time environmental data, including wind speed, irradiance, blade rotation speed, and ambient temperature. Because this data directly reflects the plant's equipment performance and its sensitivity to geographical and meteorological conditions, it constitutes core production privacy, and the plants are unwilling to upload the raw plaintext to the dispatch center. The dispatch center is responsible for collecting the anonymized features and corresponding historical power generation (Labels) from each plant, and using Gradient Boosting Decision Tree (GBDT) or similar tree models to construct a province-wide power load forecasting model to achieve more accurate dispatch optimization.
[0044] Step 2: Build and train the power load prediction model.
[0045] Specifically, its core lies in constructing a family of data transformation functions driven by a local key, and the specific steps are as follows:
[0046] (1) Joint Initialization and Global Binning: Each participant (Data Owner) and the coordinator (Coordinator) determines the feature engineering rules. For the data corresponding to each feature dimension... Each participant calculates its segment number locally and divides it into segments. each interval ,in For hyperparameters, , These are the minimum and maximum values in the sample space. Recommended value ( (the number of samples under this feature) to ensure that each interval The data density within them is basically the same.
[0047] (2) Random topological permutation of logical space:
[0048] i. Generate a random permutation vector independently for each feature dimension. ,Right now A random arrangement;
[0049] ii. Define a virtual mapping target space ,in This is a set virtual maximum value. The target space will be mapped. Divided into A logical subspace .
[0050] iii. Establish mapping relationships That is, the first feature in the original space. The elements of the nth interval will be mapped to the nth interval in the logical space specified by the permutation vector. Each segment.
[0051] This step disrupts the macroscopic probability density distribution of features by changing the global topological order of the feature values.
[0052] (3) Construction of random monotonic splines: In each permuted interval Internally, construct a strictly monotonically increasing nonlinear spline (transformation) function. To eliminate the numerical spacing characteristics, a square integral is used:
[0053]
[0054] in, The random coefficients derived from the local keys of the participants determine the curvature shape of the curve within this segment; This is the logical segment offset, derived from the previous... The cumulative width of each logical segment determines whether the mapped value falls within the range of logical segments. Within the range; It is a gain factor used to balance the numerical magnitudes between different segments.
[0055] Then, a spline mapping transformation is performed, which maps all values of each feature column onto a virtual space through the corresponding transformation function.
[0056] (4) Cascaded order-preserving encryption: For each spline mapped value Further lightweight order-preserving encryption is performed:
[0057]
[0058] in, These are the eigenvalues after transformation by a nonlinear spline function. The key generated for this column of features by the participation policy. For any symmetric order-preserving encryption algorithm, such as mOPE, ROPF, etc.
[0059] This step aims to further mask the continuity of the derivative (slope) of the spline function at the piecewise boundaries through nonlinear stretching, preventing attackers from reconstructing the spline parameters through second-order derivative analysis and thus defending against "gradient attacks".
[0060] (5) Centralized training: Each participant will provide the encrypted feature set Send to the coordinator. The coordinator directly constructs a decision tree on the encrypted domain. Two samples. Belonging to the same original interval and Based on monotonicity If they belong to different intervals, their order relation is determined by random permutation. The decision tree can capture the new order structure after the permutation, even though the global order is broken, through recursive partitioning.
[0061] This publicly available experiment demonstrates that the training method has minimal impact on the convergence and prediction accuracy of models such as CART, GBDT, and Random Forest, and does not require encryption / decryption operations during training.
[0062] As one embodiment, this disclosure takes the joint power prediction modeling between the provincial dispatch center and various new energy power stations (wind farms, photovoltaic power stations) as an example:
[0063] Data Provider (New Energy Power Plant): The power plant possesses local real-time environmental characteristic data (such as wind speed, irradiance, blade rotation speed, ambient temperature, etc.). Because this data directly reflects the power plant's equipment performance and geographical and meteorological sensitivity, and constitutes core production privacy, the power plant is unwilling to upload the original plaintext data to the dispatch center.
[0064] Data Requester (Dispatch Center): The dispatch center is responsible for collecting the anonymized features and corresponding historical power generation (Labels) of each power station, and using Gradient Boosting Decision Tree (GBDT) or similar tree models to construct a province-wide power load forecasting model to achieve more accurate dispatch optimization. The specific implementation steps are as follows:
[0065] Step 1: Adaptive Segmentation: A wind farm processes its core feature, "historical wind speed." Based on the distribution curve of this feature over a year at the farm, it is divided into segments using the equal-frequency quantile method. For example, the wind speed range of [0 m / s, 25 m / s] is divided into 100 smaller buckets to ensure that the sample size in each bucket is balanced and to eliminate the isolation of extreme weather data points in the distribution.
[0066] Step 2: Topology Permutation: Generate random permutation vectors locally at the site. The second interval, originally located in the low wind speed zone ([0.5, 0.8] m / s), may be mapped to the 87th logical segment of the target value space after permutation. This means that in the uploaded data, the low wind speed feature is numerically represented as "extremely high value," thus masking the original monotonic logic between wind speed and power.
[0067] Step 3: Nonlinear Spline Transformation: Within each permuted logic segment, the station calls the parameters derived from the local key. Construct a square integral spline function. For example, within logic segment 87, the mapping function exhibits a highly nonlinear behavior. The distribution is symmetrical. This causes the tiny observation errors, which were originally between 0.5 m / s and 0.8 m / s, to be randomly stretched or compressed after mapping, completely erasing the statistical distance characteristics of the eigenvalues.
[0068] Step 4: Cascaded Encryption and Joint Modeling: The field station performs order-preserving encryption on the transformed values and sends them to the dispatch center. The dispatch center then aggregates the encrypted features from these encrypted features with those from other field stations.
[0069] During training, the decision tree model in this public scheduling center finds the optimal gain point through recursive splitting. Taking the most common CART decision tree as an example, the algorithm traverses all possible splitting thresholds for each feature. , will the current node dataset Divided into left subset (Eigenvalue less than) (the set of) and the right subset (Eigenvalue greater than) (The set of data). For classification tasks, it selects the optimal splitting features and optimal splitting points by minimizing the Gini index:
[0070]
[0071] in, This indicates the calculation of the Gini coefficient. This represents a split point for a feature. After finding the optimal split point, the algorithm recursively performs this process on the two subtrees it splits out, until a stopping condition is met (such as reaching the maximum depth or the number of node samples falling below a threshold).
[0072] The privacy protection scheme and decision tree training mechanism proposed in this disclosure are naturally compatible, mainly in the following two dimensions:
[0073] (1) Order relation consistency: When the CART algorithm finds the optimal split point, it essentially only uses the sorting information of the feature values, without relying on the absolute Euclidean distance between the values.
[0074] (2) Robustness of topology: Although the present invention introduces random topological permutation to break global continuity, this is only equivalent to “position rearrangement” in the feature space. Since the decision tree has the characteristic of multi-level splitting, when the original features cross multiple permutation intervals, the decision tree training algorithm can automatically identify and extract these permuted logical blocks in subsequent child nodes by increasing the depth of the tree.
[0075] In summary, the model can still accurately capture the contribution of wind speed changes to power fluctuations, thus producing prediction accuracy in the encrypted domain that is almost identical to that in the plaintext domain. Meanwhile, even if the dispatch center is hacked and data is leaked, the attackers only observe a "pseudo-feature distribution" that has been randomly rearranged and nonlinearly distorted. They cannot reconstruct the actual geographical environmental parameters or equipment efficiency curves of the substation through reverse engineering, effectively ensuring the security of power industry data.
[0076] Example 2
[0077] One embodiment of this disclosure provides a power prediction system for new energy power plants based on a vertical federated decision tree, including:
[0078] The data acquisition module is used to acquire real-time environmental characteristic data;
[0079] The load forecasting module is used to input environmental characteristic data into the power load forecasting model and output the predicted power value.
[0080] The power load forecasting model is a vertical federated decision tree model. The training process of the vertical federated decision tree model is as follows: each participant generates a key and hyperparameters for each column of features according to security requirements, and generates a random topology mapping for each column of features. A nonlinear transformation function is constructed based on each mapping relationship. The generated nonlinear transformation function and lightweight order-preserving encryption are used to cascade encrypted data. The encrypted data is concentrated in the coordinator for encrypted state model training. After training is completed, the vertical federated decision tree model is obtained.
[0081] Example 3
[0082] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for predicting the power of new energy power plants based on a vertical federated decision tree.
[0083] Example 4
[0084] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When these computer instructions are executed by a processor, they implement the aforementioned new energy power prediction method based on a vertical federated decision tree.
[0085] Example 5
[0086] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned new energy power plant power prediction method based on a vertical federated decision tree.
[0087] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0088] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0089] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.
Claims
1. A power prediction method for new energy power plants based on vertical federated decision trees, characterized in that, include: Acquire real-time environmental feature data; Environmental characteristic data is input into the power load prediction model, and the predicted power value is output. The power load prediction model is a vertical federated decision tree model. The training process of the vertical federated decision tree model is as follows: each participant generates a key and hyperparameters for each column of features according to security requirements, and generates a random topology mapping for each column of features. A nonlinear transformation function is constructed based on each mapping relationship. The generated nonlinear transformation function and lightweight order-preserving encryption are used to concatenate encrypted data. The encrypted data is concentrated in the coordinator for encrypted state model training. After training is completed, the vertical federated decision tree model is obtained. For each feature dimension, an independent random permutation vector is generated. A virtual mapping target space is defined, and the mapping target space is divided into... K Establish a mapping relationship in the logical subspace, that is, the in the original space. i The elements of each interval will be mapped to the segment in the logical space specified by the permutation vector; Within each permuted logical subspace, a strictly monotonically increasing nonlinear transformation function is constructed. To eliminate numerical spacing characteristics, the nonlinear transformation function adopts a square integral form: in, The random coefficients derived from the local keys of the participants determine the curvature shape of the curve within this segment; This is the logical segment offset, derived from the previous... The cumulative width of each logical segment determines whether the mapped value falls within the logical subspace. Within the range; This is a gain factor used to balance the numerical magnitudes between different segments; The data is within the range.
2. The power prediction method for new energy power plants based on vertical federated decision trees as described in claim 1, characterized in that, Each participating party and the coordinator determine the feature engineering rules. For each feature dimension corresponding to data X, each participating party calculates the number of segments locally and divides it into segments. K Each interval is defined, and keys and hyperparameters are generated for each feature column based on security requirements.
3. The power prediction method for new energy power plants based on vertical federated decision trees as described in claim 1, characterized in that, Lightweight order-preserving encryption is applied to the values of each spline after mapping. Nonlinear stretching is used to mask the continuity of derivatives that may exist at the segment boundaries of the spline function, preventing attackers from reconstructing the spline parameters through second-order derivative analysis.
4. The power prediction method for new energy power plants based on vertical federated decision trees as described in claim 1, characterized in that, Each participant sends the encrypted feature set to the coordinator, who directly constructs a decision tree on the ciphertext field to arrange the order relations. Even though the global order is broken, the decision tree captures the new order structure after the permutation through recursive splitting.
5. A power prediction system for new energy power plants based on a vertical federated decision tree, characterized in that, include: The data acquisition module is used to acquire real-time environmental characteristic data; The load forecasting module is used to input environmental characteristic data into the power load forecasting model and output the predicted power value. The power load prediction model is a vertical federated decision tree model. The training process of the vertical federated decision tree model is as follows: each participant generates a key and hyperparameters for each column of features according to security requirements, and generates a random topology mapping for each column of features. A nonlinear transformation function is constructed based on each mapping relationship. The generated nonlinear transformation function and lightweight order-preserving encryption are used to concatenate encrypted data. The encrypted data is concentrated in the coordinator for encrypted state model training. After training is completed, the vertical federated decision tree model is obtained. Each feature independently generates a random permutation vector, defining a virtual mapping target space, which is then divided into... K Establish a mapping relationship in the logical subspace, that is, the in the original space. i The elements of each interval will be mapped to the segment in the logical space specified by the permutation vector; Within each permuted logical subspace, a strictly monotonically increasing nonlinear transformation function is constructed. To eliminate numerical spacing characteristics, the nonlinear transformation function adopts a square integral form: in, The random coefficients derived from the local keys of the participants determine the curvature shape of the curve within this segment; This is the logical segment offset, derived from the previous... The cumulative width of each logical segment determines whether the mapped value falls within the logical subspace. Within the range; This is a gain factor used to balance the numerical magnitudes between different segments; The data is within the range.
6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the power prediction method for new energy power plants based on vertical federated decision trees as described in any one of claims 1-4.
7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the power prediction method for new energy power plants based on a vertical federated decision tree as described in any one of claims 1-4.
8. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the power prediction method for new energy power plants based on a vertical federated decision tree as described in any one of claims 1-4.