New energy system development path prediction method, equipment and medium

Abstract

The invention relates to a new energy system development path prediction method and device and a medium, and the method comprises the following steps: obtaining time sequence multi-source data related to a new energy system, carrying out the data coding, and obtaining a node embedding vector; calculating a first difference factor between the node embedding vectors, and obtaining a distribution vector; calculating a second difference factor of each node embedding vector relative to the distribution vector, performing feature processing on the node embedding vectors based on the second difference factor, inputting the node embedding vectors into a multi-task learning model, and outputting a resource adaptation threshold value and a policy sensitivity threshold value; and screening the dominant nodes and the important nodes as starting points, determining a target development node as an end point, and determining an optimal path from the starting points to the end point by taking minimization of technical risks and maximization of resource utilization efficiency as targets, thereby realizing development path prediction of the new energy system. Compared with the prior art, the method has the advantages that the feature association in the node embedding vector is fully mined, and the accuracy of development path prediction can be improved.

Description

Technical Field

[0001] This invention relates to the field of development path prediction technology, and in particular to a method, device and medium for predicting the development path of a new energy system. Background Technology

[0002] Predicting the development path of new energy systems is crucial for promoting energy transition and optimizing the energy structure. Existing methods for predicting the development path of new energy systems mainly focus on the analysis and prediction of single or a few key indicators, such as power generation resource structure and flexibility resource structure. For example, by analyzing historical data and future development trends of key indicators, the evolution path of new energy systems is predicted. While these methods can provide guidance for the development of new energy systems to some extent, they struggle to consider various factors, resulting in evolution paths that are difficult to apply in practice.

[0003] CN119940638A discloses a method and apparatus for predicting industrial development paths based on knowledge graphs. It first acquires a set of economic development indicators for the region to be analyzed, and then introduces a deep learning-based data processing algorithm to perform semantic embedding encoding, hub feature analysis, and sparsification processing on each economic development indicator to uncover the overall characteristics and industrial distribution patterns of the region's economy. Subsequently, a feedforward neural network model is used to dynamically set thresholds and screen out the region's main industries. Based on this, the development path from the main industries to the target industries is predicted. This approach fully considers the diversity of economic development levels and industrial structures in different regions, dynamically adapting to the industrial characteristics and development trends of each region, thereby improving the accuracy of industrial development path prediction.

[0004] However, existing technologies typically focus on analyzing a single data source or a few types of data when processing data, neglecting the correlation and temporal characteristics between multiple data sources. This results in the inability to fully utilize rich data resources to improve the accuracy of predictions when determining thresholds, and the determined thresholds also consider only one dimension, failing to accurately reflect the actual operating status and development trend of the system. Summary of the Invention

[0005] The purpose of this invention is to overcome the defects of the existing technology and provide a method, device and medium for predicting the development path of a new energy system.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] According to a first aspect of the present invention, a method for predicting the development path of a new energy system is provided, the method comprising the following steps:

[0008] Acquire time-series multi-source data related to the new energy system and encode the data to obtain node embedding vectors;

[0009] Calculate the first difference factor among the node embedding vectors and use it as the weight distribution to perform weighted aggregation of the node embedding vectors to obtain the distribution vector; calculate the second difference factor of each node embedding vector relative to the distribution vector; perform feature processing on the node embedding vector based on the second difference factor, input it into the multi-task learning model, and output the resource adaptation threshold and policy sensitivity threshold.

[0010] Based on the resource adaptation threshold, dominant nodes are selected, and important nodes are selected based on the policy sensitivity threshold. The dominant or important nodes are used as the starting point, and the target development node is determined as the end point. With the goal of minimizing technical risks and maximizing resource utilization efficiency, the optimal path from the starting point to the end point is determined, so as to realize the prediction of the development path of the new energy system.

[0011] As a preferred technical solution, the time-series multi-source data includes technical data, resource data, policy data, and related data; the technical data includes parameters and technology iteration speed of different new energy technology nodes; the resource data includes spatial resource distribution and exploitable resource quantity; the policy data includes policy text, policy intensity, and policy implementation time; and the related data includes economic data and environmental data.

[0012] As a preferred technical solution, the method for obtaining the node embedding vector is as follows:

[0013] Each component in the new energy system is treated as a node, and the state of each type of node at different times is regarded as a different node instance.

[0014] Temporal association edges are constructed based on the temporal relationships between different instances of the same node, and association relationship edges are constructed based on the physical connections or logical relationships between different nodes, and the edge weights are determined.

[0015] Features of nodes or edges are extracted from the acquired time-series multi-source data to construct a dynamic time-series graph.

[0016] The uncertainty of node relationships in a dynamic time series graph is quantified to obtain an uncertainty measure for each edge, and the edge weights are adjusted based on the uncertainty measure.

[0017] Based on the dynamic time series graph after edge weight adjustment, cluster analysis is performed on the set of neighbor nodes of each node to identify the local distribution pattern of the node, and the aggregation weight is dynamically adjusted based on the local distribution pattern.

[0018] The embedding vector of each node instance is initialized based on the node's features, and the information of neighboring nodes is aggregated according to the aggregation weight and edge weight to obtain the final node embedding vector.

[0019] As a preferred technical solution, the first difference factor between the embedding vectors of the computing nodes, and using it as the weight distribution, is specifically as follows:

[0020] The node embedding vector is mapped to a quantum state, and the outer product of the quantum states is calculated to obtain the node density matrix;

[0021] The quantum information entropy of a node is calculated based on its density matrix, and the difference in quantum information entropy between two nodes is used as the first difference factor between the nodes.

[0022] The weights between nodes are determined by using the first difference factor as the index and combining it with the weight decay rate control parameter.

[0023] As a preferred technical solution, the calculation of the second difference factor of each node embedding vector relative to the distribution vector, and the feature processing of the node embedding vector based on the second difference factor, specifically involves:

[0024] Map the node embedding vector to the quantum state;

[0025] By introducing the interaction between the environmental quantum state and the node quantum state, and simulating the quantum decoherence process, a new quantum state is obtained;

[0026] The difference between the quantum states before and after decoherence is calculated as the second difference factor;

[0027] The node embedding vectors are sorted and filtered based on the second difference factor;

[0028] The new quantum state corresponding to the selected node embedding vector is transformed into a feature vector, which is then used as the reconstructed node embedding vector.

[0029] As a preferred technical solution, the multi-task learning model includes a shared feature extraction layer, a task-specific layer, and an inter-task interaction layer. The shared feature extraction layer extracts shared features from the node embedding vectors. The task-specific layer includes a resource adaptation threshold prediction branch and a policy sensitivity threshold prediction branch, which process the shared features and interaction features from different perspectives to obtain corresponding task feature vectors. These vectors are then output as resource adaptation thresholds and policy sensitivity thresholds through a multilayer perceptron. The inter-task interaction layer concatenates the task feature vectors from different branches to obtain a fusion vector. The fusion vector is then subjected to linear transformation and nonlinear activation to obtain the interaction features.

[0030] As a preferred technical solution, the path is a sequence of nodes, where adjacent nodes satisfy technical feasibility constraints, resource constraints, and policy compatibility constraints.

[0031] The technical feasibility constraints are implemented in the following way: Based on the equipment characteristics and operating standards of the new energy system, a knowledge base containing technical connection rules between various equipment is established. The knowledge base is traversed at each expansion node. If a corresponding rule is matched, it indicates that the technical feasibility constraints are met.

[0032] The resource constraints are achieved by constructing a resource model, evaluating the resource demand and supply capacity of each node in the path, and indicating that the resource constraints are met when the resource demand of all nodes after the expansion of nodes is less than the supply capacity.

[0033] The policy compatibility constraint is implemented in the following way: at each expansion node, a policy impact assessment is performed and policy risks are predicted. When the policy risk is less than a preset threshold, it indicates that the policy compatibility constraint is met.

[0034] As a preferred technical solution, the optimal path is obtained by solving the Pareto optimal solution using a multi-objective optimization algorithm, outputting candidate paths, and then combining expert experience to select the final optimal path. In the process of iteration, the multi-objective optimization algorithm updates the corresponding weight of each objective in the objective function according to the rate of decrease of the training loss of each objective.

[0035] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.

[0036] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. This invention acquires and encodes time-series multi-source data related to new energy systems, obtains node embedding vectors, and calculates difference factors based on these vectors. This fully mines the potential information in the multi-source data, improving the comprehensiveness and accuracy of feature representation. This provides a richer information foundation for subsequent threshold determination and path planning, solving the problem of insufficient multi-source data mining in existing technologies.

[0039] 2. This invention dynamically determines resource adaptation thresholds and policy sensitivity thresholds by calculating the difference factor between node embedding vectors. This method can adjust the thresholds in real time based on the dynamic characteristics and complex relationships between nodes in the new energy system, thereby more accurately reflecting the actual operating status and development trend of the system, overcoming the static and singular shortcomings of existing threshold determination methods. Furthermore, by performing feature processing on the node embedding vectors through a second difference factor and inputting it into a multi-task learning model to predict the policy sensitivity threshold, it can better capture the dynamic changes and uncertainties of policy factors. This allows the determined thresholds to reflect changes in the policy environment in a timely manner, enhancing the stability and adaptability of path prediction, and overcoming the shortcomings of existing technologies in insufficiently considering the dynamic impact of policy factors.

[0040] 3. This invention employs a multi-objective optimization algorithm, comprehensively considering both minimizing technological risk and maximizing resource utilization efficiency, to output a Pareto optimal solution. This enables a balance to be achieved among multiple conflicting objectives when determining resource suitability thresholds and policy sensitivity thresholds, providing a more comprehensive and reliable path planning scheme and overcoming the shortcomings of existing technologies in multi-objective optimization. Attached Figure Description

[0041] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0043] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0044] Example 1

[0045] This embodiment provides a method for predicting the development path of a new energy system, such as Figure 1 As shown, the method includes the following steps:

[0046] S1: Obtain time-series multi-source data related to the new energy system and encode the data to obtain node embedding vectors.

[0047] In this embodiment, the time-series multi-source data includes technical data, resource data, policy data, and related data; wherein,

[0048] Technical data includes: parameters of different new energy technology nodes (such as the conversion efficiency of photovoltaic / wind power, the charging and discharging efficiency of energy storage technology, cost curves, and technology maturity level (TRL)) and the speed of technology iteration (such as the annual cost reduction rate);

[0049] Resource data includes: spatial resource distribution (such as solar irradiance, wind speed, and hydropower reserves in various regions) and exploitable resources (such as usable land area and grid access capacity);

[0050] Policy data includes: policy text (such as subsidy policies, carbon emission regulations, grid connection standards), policy strength (such as subsidy amount, tax incentive ratio), and policy implementation time.

[0051] Related data include: economic data (such as investment scale and market demand) and environmental data (such as carbon emission coefficients and ecological constraints).

[0052] Existing technologies, when constructing graph models of new energy systems, often focus on static connections, neglecting the time-varying characteristics of nodes. This embodiment integrates the time-series operational data of the new energy system into the graph structure to construct a dynamic time-series graph, thereby obtaining node embedding vectors. Specifically, it includes the following steps:

[0053] S11 treats each component of the new energy system (such as power generation equipment, substations, users, etc.) as nodes, and regards the state of each type of node at different times as different node instances; for example, the power output and operating status of the power generation equipment node at different times are regarded as different node instances.

[0054] S12, construct temporal association edges based on the temporal relationship between different instances of the same node, construct association relationship edges based on the physical connection or logical relationship between different nodes (such as power transmission, policy influence, etc.), and determine the edge weights.

[0055] S13. Based on the acquired time-series multi-source data, extract the features of the constructed nodes or edges to construct a dynamic time-series graph.

[0056] Specifically, technical data, resource data, policy data, and related data are incorporated into the dynamic graph as features of nodes or edges. For example, technical parameters and data of power generation equipment are used as node features, while policy data is used as edge features. By constructing a dynamic time-series graph, the changing characteristics of nodes over time are integrated into the graph structure, preparing for the subsequent acquisition of embedding vectors that reflect the temporal characteristics of nodes.

[0057] S14, quantify the uncertainty of node relationships in the dynamic time series graph to obtain the uncertainty measure of each edge, and adjust the edge weights based on the uncertainty measure.

[0058] The purpose of this step is to quantify the uncertainty of node relationships in the dynamic time series graph so that different weights can be assigned in the subsequent embedding vector generation, thereby enhancing the robustness of the model.

[0059] Uncertainty can be assessed based on data quality, measurement errors, and external environmental factors (such as the impact of policy changes on nodes and the impact of weather changes on power generation equipment). In this embodiment, the uncertainty is calculated as follows:

[0060] u i→j =Uncertainty(D i,j Q i,j E i,j )

[0061] Among them, D i,j Q represents the difference in data quality between nodes i and j. i,j E represents the measurement error between nodes i and j. i,j The influence of external environmental factors on nodes i and j is represented by Uncertainty, which is the uncertainty measure function. i→j This represents the uncertainty measure between nodes i and j.

[0062] Subsequently, based on the uncertainty measure u i→j Adjust edge weights w i→j :

[0063]

[0064] By quantifying uncertainty, we can provide a basis for strengthening node relationships in the subsequent graph embedding process, so that the embedding vector can more accurately reflect the real relationship between nodes.

[0065] S15. Based on the dynamic time series graph after edge weight adjustment, perform cluster analysis on the set of neighboring nodes of each node to identify the local distribution pattern of the node, and dynamically adjust the aggregation weight based on the local distribution pattern.

[0066] The purpose of this step is to dynamically adjust the generation of the embedding vector based on the local distribution characteristics of the nodes, optimizing the node embedding vector to better reflect the distribution characteristics of the nodes. Specifically, it includes the following steps:

[0067] Cluster analysis is performed on the set of neighboring nodes N(i) of each node i to identify the local distribution pattern of the node.

[0068] Based on the clustering results, the parameters of the aggregation function are dynamically adjusted. For example, for nodes in densely distributed regions, an aggregation method that focuses more on subtle differences is used; for nodes in sparsely distributed regions, an aggregation method that emphasizes feature extraction is used.

[0069] In one embodiment, the local feature aggregation formula is:

[0070]

[0071] Where σ is the activation function, W and b are learnable parameters, and c m For cluster centers, s i,m Let be the similarity between node i and the cluster center, and k be the total number of nodes.

[0072] This step adaptively mines the distribution characteristics of nodes and dynamically adjusts the generation process of embedding vectors, so that the generated embedding vectors can better reflect the local and global distribution characteristics of nodes.

[0073] S16: Initialize the embedding vector of each node instance based on the node's features, and aggregate the information of neighboring nodes according to the aggregation weight and edge weight to obtain the final node embedding vector.

[0074] This step specifically includes:

[0075] Initialize the embedding vector: Initialize the embedding vector for each node instance. In this embodiment, the initialization is based on the features of the node, but in other embodiments, it can also be initialized randomly.

[0076] Aggregate neighbor information: Aggregate information about neighboring nodes based on edge weights and local distribution characteristics of nodes. During the aggregation process, the aggregation weights are dynamically adjusted to reflect the uncertainties and local distribution characteristics between nodes.

[0077] The formula for updating the embedding vector is as follows:

[0078]

[0079] in, Let W be the embedding vector of node i at layer l. (l) and b (l) w are the learnable parameters of the l-th layer. i→j The edge weights are adjusted based on uncertainty quantification and local distribution characteristics.

[0080] After multiple layers of aggregation and optimization, the final node embedding vector is obtained.

[0081] Through the above steps, embedding vectors that reflect the temporal, uncertainty, and distribution characteristics of nodes in the new energy system can be generated, providing more accurate feature representations for subsequent prediction of the development path of the new energy system.

[0082] S2, calculate the first difference factor between node embedding vectors, and use it as the weight distribution to perform weighted aggregation of node embedding vectors to obtain a distribution vector; calculate the second difference factor of each node embedding vector relative to the distribution vector, perform feature processing on the node embedding vector based on the second difference factor, input it into the multi-task learning model, and output the resource adaptation threshold and policy sensitivity threshold.

[0083] S21, Calculate the first difference factor

[0084] The node embedding vector is normalized to conform to the probability amplitude requirements of the quantum state. The normalized vector is denoted as |ψ|. i > represents the quantum state of node i:

[0085]

[0086] Among them, h i,jLet d represent the j-th component of the embedding vector of node i, where d is the dimension of the embedding vector.

[0087] The density matrix is ​​constructed using the normalized node embedding vectors. For a pure-state quantum system, the density matrix can be expressed as the outer product of the quantum states:

[0088] ρ i =|ψ i ><ψ i |

[0089] Where, ρ i The density matrix of node i, <ψ i | is the quantum state of node i|ψ i > conjugate transpose.

[0090] Calculate the quantum information entropy S(ρ) of a node based on its density matrix:

[0091] S(ρ)=-Tr(ρlogρ)

[0092] The difference in quantum information entropy between two nodes is taken as the first difference factor ΔS between the nodes. i,j :

[0093] ΔS i,j =|S(ρ i )-S(ρ j )|

[0094] S22, Obtain the distribution vector

[0095] The weights between nodes are determined by using the first difference factor as the index and combining it with the weight decay rate control parameter:

[0096]

[0097] Here, α is the weight decay rate control parameter. Unlike traditional distance-based linear weight allocation, this method is more suitable for handling complex high-dimensional quantum state differences and can better reflect the complex interactions between nodes.

[0098] Based on weight w i,j The distribution vector is obtained by weighting the node embedding vector and the node embedding vectors of its neighbors.

[0099] S23, Calculate the second difference factor

[0100] First, similar to the description in step S21, the node embedding vector is mapped to the quantum state |ψ i >

[0101] Subsequently, the environmental quantum state |∈> and the node quantum state |ψ are introduced. iInteractions, simulating quantum decoherence processes, yield new quantum states |ψ′ i >

[0102] The difference between the quantum states before and after decoherence is calculated as a second difference factor. In one embodiment, |ψ| can be calculated using metrics such as the inner product of quantum states or trace distance. i > and |ψ′ i The difference between the two vectors is used as the second difference factor. The smaller the difference, the closer the embedding vector of node i is to the distribution vector.

[0103] S24, Feature processing of node embedding vectors based on the second difference factor.

[0104] The node embedding vectors are sorted based on the second difference factor, retaining features with smaller differences and discarding features with larger differences, thus completing the feature selection.

[0105] The new quantum state corresponding to the selected node embedding vector is transformed into a feature vector, which is then used as the reconstructed node embedding vector.

[0106] S25, Multitasking Learning

[0107] Multi-task learning models aim to simultaneously predict resource suitability thresholds and policy sensitivity thresholds. While these two tasks are somewhat related, they also possess their own unique characteristics. Therefore, multi-task learning models include a shared feature extraction layer, a task-specific layer, and an inter-task interaction layer.

[0108] The shared feature extraction layer is used to extract shared features from the node embedding vectors. It can adopt structures such as multilayer perceptron (MLP) or convolutional neural network (CNN). The extracted shared features are helpful for both tasks, effectively reducing the dimensionality of the data and avoiding overfitting.

[0109] The task-specific layer includes a resource adaptation threshold prediction branch and a policy sensitivity threshold prediction branch. These branches process shared and interactive features from different perspectives to obtain corresponding task feature vectors, which are then output as resource adaptation thresholds and policy sensitivity thresholds via a multilayer perceptron. The resource adaptation threshold prediction branch focuses on resource-related features such as resource availability and demand. Combining shared and interactive features, it uses a neural network to learn and ultimately output the resource adaptation threshold. The policy sensitivity threshold prediction branch focuses on policy-related features such as policy support intensity and policy impact scope. Utilizing shared and interactive features, it uses a neural network to learn and predict the policy sensitivity threshold.

[0110] The task interaction layer concatenates the task feature vectors from different branches to obtain a fusion vector. The fusion vector is then subjected to linear transformation and nonlinear activation. The linear transformation changes the dimension and spatial structure of the features through matrix multiplication, while the nonlinear activation (such as ReLU) introduces nonlinearity to enhance the model's ability to fit complex relationships, thereby obtaining interactive features.

[0111] S3 uses a resource adaptation threshold to select leading nodes and a policy sensitivity threshold to select important nodes. It takes the leading or important nodes as the starting point and the target development node as the end point. With the goal of minimizing technical risks and maximizing resource utilization efficiency, it determines the optimal path from the starting point to the end point, thereby realizing the prediction of the development path of the new energy system.

[0112] S31, Selecting the dominant node based on resource adaptation threshold.

[0113] For each node, the resource adaptation ratio is determined based on the resource development volume and resource quality of each node.

[0114] First, for each node, assess the exploitable resources in its area, including available land area and grid connection capacity. For example, for a solar power node, calculate whether the available land area in its area is sufficient to meet the needs of large-scale photovoltaic power plant construction; for a wind power node, assess the grid connection capacity in its vicinity to determine whether it can accommodate the connection and transmission of wind power.

[0115] Secondly, resource quality scores are calculated based on the resource quality of the region where the node is located, such as solar irradiance, wind speed, and hydropower reserves. Industry standards or expert experience can be used to set scores corresponding to different resource quality levels. For example, for solar irradiance, the region can be divided into high irradiance areas (such as Tibet and Qinghai, which have higher scores), medium irradiance areas, and low irradiance areas.

[0116] Finally, based on the assessment results of resource exploitability and resource quality, the resource matching ratio is calculated according to certain weights. The calculation formula is as follows:

[0117] Resource adaptation ratio = ω1 × resource exploitability score + ω2 × resource quality score

[0118] Wherein, ω1 and ω2 are the weights of the exploitable amount of resources and the quality of resources, respectively, and ω1+ω2=1.

[0119] By comparing the resource adaptation ratio with the resource adaptation threshold, nodes with high resource adaptation are selected as dominant nodes. Methods such as threshold comparison and ranking can be used to determine dominant nodes that have a resource adaptation ratio higher than the threshold or rank within a certain proportion in the ranking.

[0120] Nodes with high resource matching ratios typically possess abundant resources and favorable development conditions, enabling them to efficiently utilize resources for new energy production. Selecting these nodes as leading nodes ensures that the core power generation nodes of the new energy system have stable and efficient resource utilization capabilities, thereby improving the overall power generation efficiency and economics of the system.

[0121] S32, Selecting Important Nodes Based on Policy Sensitivity Thresholds

[0122] For each node, a policy sensitivity score is calculated based on the strength of policy support and policy stability.

[0123] First, analyze the level of policy support for new energy projects in the regions where the nodes are located, such as subsidies, tax incentives, and land policies. For example, some regions offer higher subsidies for solar power projects or tax breaks for wind power projects; these policy supports can be converted into points.

[0124] Secondly, considering the stability and continuity of policies, the impact of policy changes on nodes is assessed. Regions with stable policies score higher, while regions with frequent policy changes score lower.

[0125] Finally, considering both the strength of policy support and policy stability, a policy sensitivity score is calculated. The calculation formula is as follows:

[0126] Policy sensitivity score = β1 × policy support strength score + β2 × policy stability score

[0127] Here, β1 and β2 are the weights of policy support strength and policy stability, respectively, and β1+β2=1.

[0128] By comparing policy sensitivity scores with policy sensitivity thresholds, nodes with high policy sensitivity are selected as important nodes. Methods such as threshold comparison and cluster analysis can be used to identify nodes with policy sensitivity scores above the threshold or those belonging to specific clusters as important nodes.

[0129] Key policy-sensitive nodes are more attuned to policy changes and better able to adapt to policy directions and utilize policy support. Identifying these nodes as critical can ensure that the development direction of the new energy system aligns with policy objectives, reduce policy risks, and enable timely adjustments to strategies to respond to policy changes, thereby enhancing the system's adaptability and sustainable development capabilities.

[0130] S33, Development Path Prediction

[0131] Starting from the dominant or important node and determining the target development node as the endpoint, find the optimal path from the starting point to the endpoint.

[0132] In this embodiment, the selection of target development nodes requires comprehensive consideration of resources, policies, market conditions, and grid conditions. First, the exploitable resources and quality of each region are assessed, such as solar irradiance, wind speed, and hydropower reserves, as well as the available land area and grid connection capacity. Next, policy guidance is analyzed, focusing on the strength and stability of government policy support for new energy projects. Then, market demand is considered, prioritizing nodes located near load centers. Finally, grid connection conditions are evaluated to ensure nodes have favorable access conditions. Based on these factors, a comprehensive score for each node is calculated and ranked, with nodes boasting high comprehensive scores selected as target development nodes. Simultaneously, the selection results are verified and adjusted using expert experience and field research to ensure the scientific validity and feasibility of the target development node selection.

[0133] The path proposed in this embodiment is a node sequence, and adjacent nodes in the path satisfy technical feasibility constraints, resource constraints, and policy compatibility constraints.

[0134] Technical feasibility constraints refer to the requirement that connections between adjacent nodes in a path must be technically feasible, meaning that stable and reliable energy transmission, conversion, or interaction between nodes must be technically possible. For example, when different types of new energy power generation equipment (such as photovoltaic power plants and wind farms) are connected to the grid, factors such as the stability of their output power, power quality, and impact on the grid need to be considered; and the connection between substations and transmission lines needs to meet technical requirements such as voltage levels and transmission capacity.

[0135] Technical feasibility constraints are implemented in the following way: Based on the equipment characteristics and operating standards of the new energy system, a knowledge base containing technical connection rules between various equipment is established. When expanding a node, the knowledge base is traversed. If a corresponding rule is matched, it indicates that the technical feasibility constraint is met, and the expanded node can be used as one of the nodes on the candidate path. Otherwise, it indicates that the feasibility constraint is not met, and the next expanded node is searched.

[0136] Resource constraints refer to the need for connections between adjacent nodes in a path to maintain a balance between resource supply and consumption, including energy resources, equipment resources, and human resources. For example, at renewable energy generation nodes, it is necessary to ensure a sufficient supply of renewable energy (such as solar and wind power); at equipment installation and maintenance nodes, the availability of equipment and the investment of maintenance resources need to be considered.

[0137] Resource constraints are implemented as follows: a resource model is constructed, and the resource demand and supply capacity of each node in the path are evaluated. When the resource demand of all nodes after the expansion node is less than the supply capacity, it indicates that the resource constraint is met, and the expansion node can be used as one of the candidate nodes in the path. Otherwise, it indicates that the resource constraint is not met, and the next expansion node is searched.

[0138] Policy compatibility constraints refer to the requirement that connections between adjacent nodes in a path must comply with current energy policies, environmental regulations, and industrial policies. For example, some regions may have specific policy support or restrictions on the construction of new energy projects, such as subsidy policies, entry thresholds, and environmental protection requirements.

[0139] Policy compatibility constraints are implemented as follows: At each node expansion, a policy impact assessment is performed, and policy risks are predicted. If the policy risk is less than a preset threshold, the policy compatibility constraint is met, and the expanded node can be considered as a candidate path node. Otherwise, the resource constraint is not met, and the next expanded node is sought. In one embodiment, policy risk prediction can employ machine learning methods. Historical policy data, project data, market data, etc., are used as training samples. Project type, policy type, time characteristics, and regional characteristics are extracted as model input variables. Classification algorithms (such as decision trees, support vector machines, neural networks, etc.) or regression algorithms in supervised learning are used to predict the policy risk level or quantify the degree of policy risk.

[0140] Using minimizing technical risks and maximizing resource utilization efficiency as objective functions, the optimal path is obtained by solving for Pareto optimal solutions using a multi-objective optimization algorithm, outputting candidate paths, and then combining expert experience to select the final optimal path.

[0141] In this embodiment, the objective function is expressed as:

[0142] L=αL1+βL2

[0143] Where α and β are the weights corresponding to the sub-objective functions L1 and L2, and L1 = min F tec L2 = max F eff F tec For technological risks, F eff For resource utilization efficiency.

[0144] In one embodiment, technological risk can be determined by comprehensively considering equipment failure rate, technology maturity, system stability, technology compatibility, and environmental adaptability. Equipment failure rate is calculated by statistically analyzing historical failure data of new energy equipment, determining the mean time between failures (MTBF) and mean time to repair (MTTR). A higher failure rate indicates greater technological risk. Technology maturity is assessed using a technology maturity model (such as TAM) to evaluate the maturity level of new energy technologies. Technologies with low maturity levels have higher uncertainty and risk. System stability is calculated using simulations and historical data to determine stability indicators of the new energy system under different operating conditions, such as frequency fluctuations and voltage deviations. Poor system stability implies high technological risk. The technological compatibility of new energy technologies with existing systems and other technologies is assessed, including technical interfaces and standard matching. Technologies with poor compatibility may bring higher integration risks. The adaptability of the new energy system to environmental factors (such as temperature, humidity, and wind speed) is considered, and the system's performance and reliability under different environmental conditions are evaluated to obtain environmental adaptability indicators. Poor environmental adaptability indicates high technological risk.

[0145] In one embodiment, resource utilization efficiency can be determined by comprehensively considering resource development efficiency, resource allocation rationality, resource conversion efficiency, and resource utilization stability. Resource development efficiency is determined by calculating the ratio of actual developed new energy resources to theoretically exploitable resources. For example, for solar power generation, this is calculated as the ratio of actual power generation to the theoretical maximum power generation. Resource allocation rationality refers to assessing whether the allocation of resources at different nodes and time periods is reasonable, and whether there is resource waste or insufficiency. Scoring can be based on the uniformity and matching degree of resource allocation. Resource conversion efficiency is obtained by calculating the efficiency of converting resources into usable energy in the new energy system, such as the photoelectric conversion efficiency of solar panels and the wind energy conversion efficiency of wind turbines. The stability and sustainability of resource utilization are assessed by calculating the volatility and reliability indicators of resource supply.

[0146] After determining the objective function, a multi-objective optimization algorithm is used to solve for the Pareto optimal solution. Algorithms that can be used include, but are not limited to, the NSGA-II algorithm, multi-objective particle swarm optimization, multi-objective differential evolution algorithm, multi-objective genetic algorithm, and multi-objective simulated annealing algorithm. Those skilled in the art, given the objective function, are free to choose a multi-objective optimization algorithm for solving the problem. To avoid ambiguity in the purpose of this application, the solution process will not be described in detail here.

[0147] After obtaining candidate paths, the final optimal path is selected by combining expert experience, thus completing the prediction of the development path of the new energy system.

[0148] Example 2

[0149] Based on Example 1, this embodiment provides a method for determining the optimal edge weight, as follows:

[0150] For nodes with frequent interactions, such as the power transmission edge between power generation equipment and substations, the interaction frequency can be statistically analyzed based on historical data. Higher frequency results in higher weight. The formula is:

[0151]

[0152] Where count(i,j) represents the frequency of interaction between nodes i and j, w i→j This represents the edge weight between nodes i and j.

[0153] For nodes with causal or dependent relationships, such as edges representing the impact of policy data on new energy project construction, the weights can be determined based on the conditional probability P, using the following formula:

[0154] w i→j =P(k|i)

[0155] For edges in a dynamic time series graph, their weights may change over time. A time decay factor, such as an exponential decay function, can be introduced.

[0156] w i→j (t)=w i→j (0)·e -αt

[0157] Among them, w i→j (0) is the initial weight, α is the decay coefficient, and t is time.

[0158] In practical applications, multiple factors can be considered to determine edge weights. For example, for an edge connecting a power generation equipment node and a policy node, the final weight can be determined by a weighted fusion method that integrates factors such as interaction frequency, conditional probability, and temporal decay.

[0159]

[0160] Here, β is a coefficient used to balance the influence of interaction frequency and conditional probability.

[0161] Example 3

[0162] This embodiment provides an improved method for a multi-objective optimization algorithm based on Embodiment 1.

[0163] Specifically, during the iteration process, the multi-objective optimization algorithm updates the corresponding weights of each objective in the objective function based on the rate of decrease of the training loss for each objective:

[0164]

[0165] in, This represents the loss value of the i-th target in the k-th iteration. Let γ represent the loss ratio of the j-th target in the k-th iteration, where γ is the temperature parameter and m is the number of targets.

[0166] Example 4

[0167] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0168] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0169] The processing unit executes the various methods and processes described above, such as methods S1 to S3. For example, in some embodiments, methods S1 to S3 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of methods S1 to S3 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S3 by any other suitable means (e.g., by means of firmware).

[0170] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0171] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0172] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0173] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the development path of a new energy system, characterized in that, The method includes the following steps: Acquire time-series multi-source data related to the new energy system and encode the data to obtain node embedding vectors; Calculate the first difference factor among the node embedding vectors and use it as the weight distribution to perform weighted aggregation of the node embedding vectors to obtain the distribution vector; calculate the second difference factor of each node embedding vector relative to the distribution vector; perform feature processing on the node embedding vector based on the second difference factor, input it into the multi-task learning model, and output the resource adaptation threshold and policy sensitivity threshold. Based on the resource adaptation threshold, dominant nodes are selected, and important nodes are selected based on the policy sensitivity threshold. The dominant or important nodes are used as the starting point, and the target development node is determined as the end point. With the goal of minimizing technical risks and maximizing resource utilization efficiency, the optimal path from the starting point to the end point is determined, so as to realize the prediction of the development path of the new energy system.

2. The method for predicting the development path of a new energy system according to claim 1, characterized in that, The aforementioned time-series multi-source data includes technical data, resource data, policy data, and related data; the technical data includes parameters and technology iteration speed of different new energy technology nodes; the resource data includes spatial resource distribution and exploitable resource quantity; the policy data includes policy text, policy intensity, and policy implementation time; and the related data includes economic data and environmental data.

3. The method for predicting the development path of a new energy system according to claim 1, characterized in that, The method for obtaining the node embedding vector is as follows: Each component in the new energy system is treated as a node, and the state of each type of node at different times is regarded as a different node instance. Temporal association edges are constructed based on the temporal relationships between different instances of the same node, and association relationship edges are constructed based on the physical connections or logical relationships between different nodes, and the edge weights are determined. Features of nodes or edges are extracted from the acquired time-series multi-source data to construct a dynamic time-series graph. The uncertainty of node relationships in a dynamic time series graph is quantified to obtain an uncertainty measure for each edge, and the edge weights are adjusted based on the uncertainty measure. Based on the dynamic time series graph after edge weight adjustment, cluster analysis is performed on the set of neighbor nodes of each node to identify the local distribution pattern of the node, and the aggregation weight is dynamically adjusted based on the local distribution pattern. The embedding vector of each node instance is initialized based on the node's features, and the information of neighboring nodes is aggregated according to the aggregation weight and edge weight to obtain the final node embedding vector.

4. The method for predicting the development path of a new energy system according to claim 1, characterized in that, The first difference factor between the node embedding vectors is used as the weight distribution, specifically: The node embedding vector is mapped to a quantum state, and the outer product of the quantum states is calculated to obtain the node density matrix; The quantum information entropy of a node is calculated based on its density matrix, and the difference in quantum information entropy between two nodes is used as the first difference factor between the nodes. The weights between nodes are determined by using the first difference factor as the index and combining it with the weight decay rate control parameter.

5. The method for predicting the development path of a new energy system according to claim 1, characterized in that, The calculation of the second difference factor of each node embedding vector relative to the distribution vector, and the feature processing of the node embedding vector based on the second difference factor, specifically involves: Map the node embedding vector to the quantum state; By introducing the interaction between the environmental quantum state and the node quantum state, and simulating the quantum decoherence process, a new quantum state is obtained; The difference between the quantum states before and after decoherence is calculated as the second difference factor; The node embedding vectors are sorted and filtered based on the second difference factor; The new quantum state corresponding to the selected node embedding vector is transformed into a feature vector, which is then used as the reconstructed node embedding vector.

6. The method for predicting the development path of a new energy system according to claim 1, characterized in that, The multi-task learning model includes a shared feature extraction layer, a task-specific layer, and an inter-task interaction layer. The shared feature extraction layer extracts shared features from the node embedding vectors. The task-specific layer includes a resource adaptation threshold prediction branch and a policy sensitivity threshold prediction branch, which process the shared features and interaction features from different perspectives to obtain corresponding task feature vectors. These vectors are then output as resource adaptation thresholds and policy sensitivity thresholds through a multilayer perceptron. The inter-task interaction layer concatenates the task feature vectors from different branches to obtain a fusion vector. The fusion vector is then subjected to linear transformation and nonlinear activation to obtain the interaction features.

7. The method for predicting the development path of a new energy system according to claim 1, characterized in that, The path is a sequence of nodes, where adjacent nodes satisfy technical feasibility constraints, resource constraints, and policy compatibility constraints. The technical feasibility constraints are implemented in the following way: Based on the equipment characteristics and operating standards of the new energy system, a knowledge base containing technical connection rules between various equipment is established. The knowledge base is traversed at each expansion node. If a corresponding rule is matched, it indicates that the technical feasibility constraints are met. The resource constraints are achieved by constructing a resource model, evaluating the resource demand and supply capacity of each node in the path, and indicating that the resource constraints are met when the resource demand of all nodes after the expansion of nodes is less than the supply capacity. The policy compatibility constraint is implemented in the following way: at each expansion node, a policy impact assessment is performed and policy risks are predicted. When the policy risk is less than a preset threshold, it indicates that the policy compatibility constraint is met.

8. The method for predicting the development path of a new energy system according to claim 1, characterized in that, The optimal path is obtained by solving the Pareto optimal solution using a multi-objective optimization algorithm, outputting candidate paths, and then combining expert experience to select the final optimal path. In the process of iteration, the multi-objective optimization algorithm updates the corresponding weight of each objective in the objective function according to the rate of decrease of the training loss of each objective.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images