Power grid investment optimization method based on big data analysis

Abstract

The invention relates to a power grid investment optimization method based on big data analysis, and belongs to the technical field of power grid investment optimization. The method comprises the following steps: accessing multi-source heterogeneous data, performing edge preprocessing and cloud fusion alignment, and storing the data to a time sequence database after standardization processing is completed; an LSTM model is constructed based on the time sequence database, a future power load and new energy power generation curve is predicted, the energy storage demand is quantified, and the project economy is evaluated; generating an optimal investment scheme set through a multi-objective optimization algorithm based on the predicted power grid state, and quantitatively evaluating the risk and income of each scheme by adopting Monte Carlo simulation; and carrying out geospatial visualization display on the investment scheme, and dynamically feeding back and optimizing power grid planning based on real-time monitoring data. According to the method, demand scheduling and energy storage scale quantification are realized through power grid state prediction, so that multi-objective optimization and risk assessment are completed, and maximization of investment benefits and multi-objective balance are realized.

Description

Technical Field

[0001] This invention belongs to the field of power grid investment optimization technology, specifically relating to a power grid investment optimization method based on big data analysis. Background Technology

[0002] Traditional power grid investment planning methods have shown significant limitations in the face of a complex environment characterized by a high proportion of renewable energy grid integration, rapid growth of new loads such as electric vehicles, and the deepening of the electricity market.

[0003] Traditional investment analysis relies on linear extrapolation from historical data and static empirical judgments. This leads to the isolation and ineffective integration of multi-source, heterogeneous power grid internal operation data, external meteorological and environmental information, user electricity consumption behavior, and market policy data, creating data silos and resulting in insufficient accuracy in predicting future power grid conditions. In particular, it fails to accurately capture the strong volatility and randomness of renewable energy output. At the same time, traditional methods often optimize based on single objectives such as cost or reliability, making it difficult to coordinate conflicts between multiple objectives such as economy, reliability, and environmental protection. Furthermore, they lack quantitative assessment methods for uncertain risks such as technical failures, electricity price fluctuations, and policy adjustments. Decisions based solely on deterministic scenarios expose investments to significant potential risks. Summary of the Invention

[0004] To address the aforementioned problems in existing technologies, this invention provides a power grid investment optimization method based on big data analysis. By predicting power grid conditions, it achieves demand scheduling and energy storage scale quantification, thereby completing multi-objective optimization and risk assessment, and maximizing investment benefits while achieving a balance between multiple objectives.

[0005] The objective of this invention can be achieved through the following technical solutions: This disclosure provides a power grid investment optimization method based on big data analysis, including the following steps: S1. Data Acquisition: Access multi-source heterogeneous data, perform edge preprocessing and cloud fusion and alignment, and store the data in a time-series database after standardization. S2. Power Grid Status Prediction: Based on the time series database, an LSTM model is built to predict future power load and new energy generation curves, thereby quantifying energy storage demand and evaluating project economics. S3. Optimization and Risk Assessment: Based on the predicted power grid state, a set of optimal investment schemes is generated through a multi-objective optimization algorithm, and Monte Carlo simulation is used to quantitatively evaluate the risks and benefits of each scheme. S4. Visualization and Feedback: Visualize the investment plan in geospatial space and dynamically optimize the power grid plan based on real-time monitoring data.

[0006] Furthermore, the data acquisition includes the following steps: S11. Multi-source data access: Systematically access various data sources, including internal power grid data, external environmental data, user and market data, and policy and planning data; S12. Data preprocessing: The collected data is preprocessed by edge computing units deployed on the edge side, including preliminary cleaning, aggregation and compression, and format conversion of high-frequency data; The cloud platform analyzes user and market data, as well as policy and planning data, and transforms them into processing data that correlates execution time with spatial relationships. S13. Data Fusion and Storage: The data preprocessed at the edge is transmitted to the cloud platform and aligned with the data parsed by the cloud platform based on timestamps and geographic information. Then, normalization is used to standardize the data of different dimensions to a unified range. All processed data is stored in a time-series database.

[0007] Furthermore, the internal power grid data includes historical and real-time load data, operating status, equipment ledgers, and fault records from SCADA and EMS systems; the external environment data includes solar radiation, wind speed, and temperature data accessed through APIs or sensor networks from meteorological bureaus / own monitoring points; the user and market data includes the number of electric vehicles, battery capacity, charging rate, and usage habits data from the charging pile operation platform, as well as electricity price information from the electricity market; and the policy and planning data includes energy policies, regulations, regional land space plans, and economic development indicators issued by governments at all levels.

[0008] Furthermore, the power grid state prediction includes the following steps: S21. Dataset Construction: Extract key features from the time series database, construct the model training dataset, and divide it into training set, validation set, and test set in chronological order; S22. Data Prediction: Using a long short-term memory network model, with the constructed dataset as input, predict the load curves and new energy power generation curves for the short and medium term. S23. Calculation of indicators: Based on the predicted power generation and load, calculate the difference at each future time point to generate a supply-demand difference curve; identify demand scheduling periods by setting power thresholds; quantify the energy storage scale based on the demand scheduling periods; and finally use NPV and IRR models to conduct a preliminary economic assessment of the proposed projects.

[0009] Furthermore, the dataset includes a load forecast dataset and a new energy power generation forecast dataset; The load forecast dataset includes time characteristics, meteorological characteristics, historical loads, and economic data; the new energy power generation forecast dataset includes historical power, wind power data, and photovoltaic data.

[0010] Furthermore, the identification of demand scheduling periods includes: When the supply-demand difference is greater than zero and greater than the power threshold, it is marked as a charging scheduling period; When the supply-demand difference is less than zero and its absolute value is greater than the power threshold, it is marked as a discharge scheduling period; When the absolute value of the supply-demand difference is less than or equal to the power threshold, it is marked as a balanced state.

[0011] Furthermore, the quantification of the energy storage scale includes: Traverse the discharge scheduling period, find the value with the largest power shortage among all time points as the power demand value, classify the discharge scheduling period into different shortage event groups, and then calculate the sum of the power shortage values ​​of all time points in each shortage event group. Select the value with the largest sum of power shortage values ​​as the energy capacity demand value.

[0012] Furthermore, the preliminary economic assessment of the proposed project using NPV and IRR models includes: By constructing a cash flow model, calculating key economic indicators, including net present value and internal rate of return, and setting a minimum economic threshold, projects can be screened. In the cash flow model, the investment cost is estimated based on energy storage power and capacity requirements, with reference to the current unit cost.

[0013] Furthermore, the optimization and risk assessment includes the following steps: S31. Optimization Problem Definition: Define decision variables, including the scale of the energy storage system, the lines to be upgraded, and the equipment selection; construct a multi-objective function, including minimizing the total investment cost, maximizing the return on investment, maximizing power supply reliability, and minimizing the overall risk level; set constraints, including the budget ceiling and physical energy storage capacity, where the constraint on physical energy storage capacity is set as the total energy storage demand minus the virtual energy storage compensation parameter; Among them, the virtual energy storage compensation parameter is the alternative energy storage capacity calculated based on electric vehicle data; S32. Multi-objective optimization solution: The genetic algorithm is used to solve the multi-objective optimization problem. First, a population representing different investment schemes is randomly initialized. Then, the merits of each scheme are evaluated through the fitness function. By simulating the natural selection process, the population is subjected to selection, crossover and mutation operations, and iterative evolution, which finally converges to a set of Pareto optimal solutions. S33. Global Risk Assessment: A global risk assessment is performed on the Pareto optimal solution set obtained through optimization. First, risk variables are identified, including technological risk, market risk, policy risk, and environmental and social risks. Then, based on the risk variables, the Monte Carlo simulation method is used to conduct random sampling simulations, calculate the probability distribution of returns for each investment option, and output the expected NPV and VaR values. S34, Sorting Label: Read the expected NPV and VaR values ​​of each scheme, plot all schemes on a two-dimensional scatter plot, and then automatically classify the schemes falling into different regions into classification groups through a clustering algorithm, and add a strategy label field to each scheme. The classification groups include: high-risk, high-return, low-risk, stable, and balanced.

[0014] Furthermore, the visualization and feedback include the following steps: S41. Geospatial Visualization: Based on the Pareto optimal solution set, all investment plans are obtained, and the geometric data in the plans are converted into standard geographic information format and plotted on a GIS map. S42. Real-time feedback and adjustment: After the investment plan is implemented, the actual operating data of the power grid is continuously monitored through IoT devices, and a deviation threshold is set for early warning. When the deviation threshold is reached, the latest actual data is used as input to re-execute the power grid state prediction.

[0015] The beneficial effects of this invention are as follows: This invention integrates and merges multi-source data from both inside and outside the power grid, and after edge preprocessing and cloud alignment and standardization, forms a high-quality time-series database, laying a solid foundation for subsequent analysis. It then uses an LSTM model to accurately predict future load and renewable energy generation curves, and quantifies energy storage demand and economic viability. Based on this, a multi-objective genetic algorithm is used to generate a Pareto optimal investment scheme set that considers cost, benefit, and reliability. Monte Carlo simulation is then used to quantify the expected return and value at risk (VaR) of each scheme under technological, market, and policy risks. Finally, a dynamic optimization closed loop is constructed through geographic information visualization and real-time data feedback mechanisms, thereby comprehensively improving the scientific nature, adaptability, and risk controllability of investment decisions. Attached Figure Description

[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 A schematic diagram illustrating the steps of a power grid investment optimization method based on big data analysis, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the data acquisition steps provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the steps of power grid state prediction provided in an embodiment of the present invention; Figure 4 A schematic diagram illustrating the optimization and risk assessment steps provided in this embodiment of the invention; Figure 5 A schematic diagram illustrating the visualization and feedback steps provided in the embodiments of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0019] This embodiment provides a power grid investment optimization method based on big data analysis, such as... Figure 1 As shown, it includes the following steps: S1. Data Acquisition: Access multi-source heterogeneous data, perform edge preprocessing and cloud fusion and alignment, complete standardization processing, and store the data in a time-series database, such as... Figure 2 As shown, it includes the following steps: S11. Multi-source data access: Systematically access various data sources, including internal power grid data, external environmental data, user and market data, and policy and planning data; The data includes: internal power grid data such as historical and real-time load data, operating status, equipment ledgers, and fault records from SCADA (Supervisory Control and Data Acquisition) and EMS (Energy Management System); external environmental data such as solar radiation, wind speed, and temperature data accessed via API or sensor networks from meteorological bureaus / own monitoring points; user and market data such as the number of electric vehicles, battery capacity, charging rate, usage habits, and electricity market pricing information from charging pile operation platforms; and policy and planning data such as energy policies, regulations, regional land use plans, and economic development indicators issued by governments at all levels.

[0020] S12. Data Preprocessing: Preprocessing of collected data is performed through edge computing units deployed at the edge, including preliminary cleaning of high-frequency data to remove obvious outliers (such as negative load values) and filling in minor missing values ​​using methods such as linear interpolation. Simultaneously, the data is aggregated and compressed to reduce network transmission pressure, and data from different protocol formats is uniformly converted to a standard format (such as JSON).

[0021] It should be noted that user and market data, as well as policy and planning data, are usually directly entered into the cloud through API interfaces, file uploads, database synchronization, etc. After structured data parsing, unstructured data parsing, and spatial data parsing, they are converted into data that can realize time-to-space correlation, such as non-time-series data (such as policy NLP technology), generating valid time ranges, and mapping policy regions to GIS geographic coordinates or administrative region codes.

[0022] S13. Data Fusion and Storage: The pre-processed data from the edge is transmitted to the cloud platform for cloud-based data fusion and storage. First, the pre-processed data from the edge and the parsed data from the cloud are aligned using timestamps and geographic information systems (GIS) to ensure consistency in the spatiotemporal dimensions. Then, Z-Score or Min-Max normalization methods are used to standardize data of different dimensions to a unified range, completely eliminating the influence of different dimensions. Finally, all processed data is stored in a time-series database, forming a high-quality, multi-dimensional historical dataset, laying a solid and reliable foundation for subsequent in-depth analysis and data mining.

[0023] S2. Power Grid Condition Prediction: Based on a time-series database, an LSTM model is constructed to predict future power load and renewable energy generation curves, thereby quantifying energy storage demand and assessing project economics. Figure 3 As shown, it includes the following steps: S21. Dataset Construction: Extract key features from the time-series database, construct the dataset for model training, and divide it into training set, validation set, and test set in chronological order.

[0024] S22. Data Prediction: Using a Long Short-Term Memory (LSTM) network model, with the constructed dataset as input, predict the load curve and new energy power generation curve for the short term (e.g., the next 7 days) and medium term (e.g., the next 1 year).

[0025] Specifically, firstly, dedicated datasets for load and renewable energy generation forecasts are selected from the dataset. The load forecast dataset includes time characteristics, meteorological characteristics, historical load, and economic data; the renewable energy generation forecast dataset includes historical power, wind power data, and photovoltaic data.

[0026] Understandably, economic data includes electricity prices and major events, wind power data includes wind speed, wind direction, and air density, and photovoltaic data includes solar irradiance, cloud cover, cloud movement, and temperature (which affects panel efficiency).

[0027] During model training, independent LSTM neural network models were built for the two prediction tasks. The networks consisted of multiple stacked LSTM units. To prevent the model from overfitting (memorizing noise from the training data), Dropout layers were inserted between the LSTM layers, randomly and temporarily ignoring a portion of neurons to enhance the model's generalization ability. The network ended with a fully connected layer responsible for mapping the complex temporal features extracted by the L units to the final predicted values. The models were compiled using the "Adam" optimizer and its mean squared error loss function, and then began iterative learning on the training data.

[0028] S23. Calculation of indicators: Based on the predicted power generation and load, calculate the difference at each future time point to generate a supply-demand difference curve; identify demand scheduling periods by setting power thresholds; quantify the energy storage scale based on the demand scheduling periods; and finally use NPV and IRR models to conduct a preliminary economic assessment of the proposed projects.

[0029] The identified demand scheduling periods include: When the supply-demand difference is greater than zero and also greater than the power threshold, it is marked as a charging scheduling period. At this time, there is a significant power surplus, which is an ideal window to guide the energy storage system to charge. When the supply-demand difference is less than zero and its absolute value is greater than the power threshold, it is marked as a discharge scheduling period. At this time, there is a significant power shortage, and the energy storage system must be required to discharge or other power sources must be required to increase their output to ensure power supply security. When the absolute value of the supply-demand difference is less than or equal to the power threshold, it is marked as a balanced state, at which point no additional energy storage scheduling intervention is required.

[0030] The quantification of energy storage scale includes: Iterate through the discharge scheduling periods and find the value with the largest power shortage at all time points (i.e., the largest absolute value among all negative values, which is the power capacity requirement that the energy storage system must meet) as the power demand value. Classify the discharge scheduling periods into different shortage event groups (for example, a continuous evening peak shortage may last for 4 hours, constituting an event group). Then, for each shortage event group, calculate the sum of the power shortage values ​​at all time points within the group (i.e., calculate the total power shortage of the entire shortage event). Select the value with the largest sum of power shortage values ​​as the energy capacity demand value.

[0031] A preliminary economic assessment of the proposed project was conducted using NPV and IRR models, including: By constructing a cash flow model, calculating key economic indicators, including net present value (NPV) and internal rate of return (IRR), and setting minimum economic thresholds (such as NPV>0 and IRR>8%), projects can be screened.

[0032] The investment cost in the cash flow model is estimated based on energy storage power and capacity requirements, referencing current unit construction costs. The calculation of key economic indicators will not be elaborated upon here.

[0033] It should be noted that the power grid state prediction uses the standardized historical data produced in the previous step to train the model to generate predictive insights into the future. Finally, the "future map" derived from the historical information—which includes predicted load, power generation, supply and demand gap and economic indicators—provides clear goals and constraints for subsequent optimization decisions.

[0034] S3. Optimization and Risk Assessment: Based on the predicted power grid state, a set of optimal investment schemes is generated using a multi-objective optimization algorithm. Monte Carlo simulation is then used to quantitatively evaluate the risks and returns of each scheme, such as... Figure 4 As shown, it includes the following steps: S31. Optimization Problem Definition: Define decision variables, including the scale of the energy storage system (capacity and power), the lines to be upgraded, and equipment selection; construct a multi-objective function, including minimizing total investment cost, maximizing return on investment, maximizing power supply reliability, and minimizing overall risk level; set constraints, including budget ceiling and physical energy storage capacity, where the constraint on physical energy storage capacity is set as total energy storage demand minus virtual energy storage compensation parameter; Among them, the virtual energy storage compensation parameter is the alternative energy storage capacity calculated based on electric vehicle (EV) data.

[0035] Understandably, the constructed objective function includes multiple conflicting objectives, such as Minimize f1(x) = total investment cost, Maximize f2(x) = total revenue (grid access fees, ancillary services, etc.), Maximize f3(x) = power supply reliability (availability, probability of load shedding (LOLP)), and Minimize f4(x) = environmental and social impact (carbon emissions, land acquisition area). Each objective function is a mathematical equation whose inputs are decision variables and data obtained from grid state prediction, including load forecast curves, new energy generation curves, power supply and demand differences, virtual energy storage compensation parameters, economic parameters (equipment unit price, operation and maintenance costs, electricity price prediction, policy subsidy standards, etc.), and physical constraints (existing grid structure, line capacity, safety regulations, available land, etc.).

[0036] The constraints are set as follows: g1(x): total cost ≤ budget limit, g2(x): physical energy storage capacity ≥ (total energy storage demand - virtual energy storage compensation parameter), g3(x): line power flow ≤ maximum line capacity (N-1 safety criterion), and decision variables >= 0 (or 0 / 1). These constraints limit the search space of the optimization algorithm, ensuring that any solution generated is engineering-feasible.

[0037] S32. Multi-objective optimization solution: The genetic algorithm (GA) is used to solve the multi-objective optimization problem. First, a population representing different investment schemes is randomly initialized. Then, the merits of each scheme are evaluated through the fitness function. Then, the natural selection process is simulated, and the population is subjected to selection, crossover, mutation and other operations. The population is iteratively evolved and finally converges to a set of Pareto optimal solutions.

[0038] It should be noted that the initialization of the GA randomly generates the first generation of the "population". Each individual is a "chromosome", representing a complete investment plan (i.e., a specific set of values ​​for a set of decision variables). For each individual (plan) in the population: decode the chromosome to obtain the specific variable values; calculate the objective function: substitute the variable values ​​into f1(x), f2(x), f3(x), f4(x) to calculate the cost, benefit, etc. of the current plan; check constraints: calculate g1(x), g2(x)... to determine whether the plan is feasible. Each plan in a Pareto optimal plan set contains complete decision variable values ​​and corresponding multiple objective function values. For example, for a solution A, if no other solution B is found that is better than A in all objectives (e.g., B has lower cost, higher benefit, and better reliability), then A is a non-dominated solution, i.e., a Pareto optimal solution. These solutions represent a set of optimal trade-offs.

[0039] S33. Global Risk Assessment: A global risk assessment is performed on the optimized Pareto optimal solution set. First, risk variables are identified, including four main categories of potential risks: technological risks (equipment failure rate, fluctuations in actual renewable energy output), market risks (future electricity prices, demand fluctuations), policy risks (subsidy phase-out time, carbon tax pricing), and environmental and social risks (public protests). Then, based on these risk variables, Monte Carlo simulations are used to conduct tens of thousands of random sampling simulations to calculate the probability distribution of returns for each investment option (e.g., "This option has an 80% probability that its IRR will exceed 8%)." Quantitative indicators such as expected NPV and Value at Risk (VaR) are output, providing decision-makers with comprehensive and quantitative risk insights.

[0040] Understandably, the Monte Carlo simulation method first sets a probability distribution function for each risk variable (e.g., electricity price may follow a normal distribution (mean = 0.65 yuan / kWh, standard deviation = 0.1)). Based on the predefined probability distribution, a set of values ​​is randomly selected for all risk variables. This set of random values ​​is substituted into the financial model to recalculate its core evaluation indicators (e.g., NPV). The NPV calculated in this simulation is stored. After 10,000 simulations, 10,000 possible NPV results are obtained, forming a probability distribution. The expected NPV is the average of the NPVs from the 10,000 simulations. Value at Risk (VaR): at a 95% confidence level, the 5th percentile of the distribution is found. For example, VaR(95%) = -100 million yuan means there is a 5% probability that the loss will exceed 100 million yuan.

[0041] S34, Sorting Label: Read the expected NPV and VaR values ​​of each scheme, plot all schemes on a two-dimensional scatter plot, and then automatically classify the schemes falling into different regions into classification groups using a clustering algorithm (such as K-Means), and add a strategy label field to each scheme.

[0042] It should be noted that in this embodiment, the ranking label is based on the grouping of the dominant strategy. Each strategy is automatically labeled according to its position in the "risk-return" quadrant. The classification groups include high-risk, high-return, low-risk, conservative, and balanced strategies. High-risk, high-return strategies are characterized by very high expected IRR and NPV, but also a large Value at Risk (VaR) (i.e., a potentially large maximum loss). Monte Carlo simulations show that their return probability distribution is very "dispersed" (high uncertainty). Low-risk, conservative strategies are characterized by moderate or even low expected IRR and NPV, but a small VaR. Their return probability distribution is very "concentrated" (high certainty), almost guaranteeing a stable return, albeit not high). Balanced strategies achieve a balance between risk and return. None of the indicators are outstanding, nor are there any obvious weaknesses.

[0043] Understandably, the final decision report generated by optimization and risk assessment includes a Pareto optimal solution set, multi-objective comparison, and risk quantification results. The Pareto optimal solution set consists of multiple candidate solutions and their detailed configurations; the multi-objective comparison compares the cost, benefits, reliability, and other multi-dimensional indicators of each solution; and the risk quantification is the probability distribution, expected value, and VaR of each solution's key financial indicators (such as NPV).

[0044] S4. Visualization and Feedback: The investment plan will be displayed in geospatial visualization, and power grid planning will be dynamically optimized based on real-time monitoring data. Figure 5 As shown, it includes the following steps: S41. Geospatial Visualization: Based on the Pareto optimal solution set, all investment plans are obtained, and the geometric data in the plans are converted into standard geographic information format and plotted on a GIS map. Understandably, the investment plan includes decision variables that require space to be displayed: Route plan: Vector data consisting of a series of latitude and longitude coordinate points; Site selection scheme: precise latitude and longitude coordinates of substations and energy storage stations; Impact range: The buffer radius required for the project (e.g., a 50-meter area on each side of the new line).

[0045] The map service overlays the solution data as an independent layer on the base map. Different solutions or different equipment types (lines, substations) are rendered using different legends (colors, line widths, icons), and a click-to-query function is provided to enable interactive functionality.

[0046] S42. Real-time feedback and adjustment: After the investment plan is implemented, the actual operating data of the power grid (load, power generation, equipment status) will be continuously monitored through Internet of Things (IoT) devices. A deviation threshold will be set (such as the actual load deviating from the predicted load by more than 15%) for early warning. When the deviation threshold is reached, the latest actual data will be used as input to re-execute the power grid status prediction.

[0047] Understandably, the input to visualization and feedback is the Pareto optimal solution set and its risk report output from the third step, and the output is an integrated, interactive decision cockpit and a continuously optimized feedback mechanism.

[0048] This invention constructs a high-quality time-series database that unifies spatiotemporal data access, edge preprocessing, and cloud integration, completely breaking down data silos and laying a solid foundation for in-depth analysis. Based on this, it utilizes an LSTM neural network model to perform deep learning on historical data, achieving high-precision predictions of short- and medium-term load curves and renewable energy generation curves. Furthermore, it derives supply-demand gaps, energy storage capacity requirements, and preliminary economic indicators for projects, enabling the prediction results to directly serve investment decisions. Moreover, based on the predicted future power grid state map, it constructs an optimization model containing multiple objective functions and constraints, including minimizing cost, maximizing revenue, and reliability, and solves it using a genetic algorithm, generating a set of Pareto-optimal investment schemes representing different trade-offs. This overcomes the limitations of single-objective optimization. Addressing future uncertainties, the solution further incorporates Monte Carlo simulation, conducting tens of thousands of random sampling simulations of various technological, market, policy, and environmental risk variables to calculate quantitative indicators such as the expected net present value (NPV) and value at risk (VaR) for each investment plan. This transforms vague qualitative risks into precise probability distributions, providing decision-makers with comprehensive risk insights. Finally, the optimal plan is spatially visualized using a Geographic Information System (GIS), and after investment execution, continuous monitoring of actual operational data is achieved through IoT devices. By setting deviation thresholds to trigger data re-collection and rolling model optimization, a dynamic closed-loop decision-making mechanism integrating "monitoring-prediction-optimization-evaluation-feedback" is formed.

[0049] This invention first constructs a dynamic database integrating heterogeneous data from multiple internal and external sources. Then, it trains an LSTM neural network model to predict future grid conditions (load and renewable energy output), thereby quantifying future supply-demand gaps and energy storage demand. Simultaneously, it uses the predicted "future map" to construct the input objectives and constraints of an investment optimization model. By defining multiple objective functions and constraints, including economic efficiency and reliability, and using a multi-objective genetic algorithm to solve them, a set of Pareto optimal investment schemes representing different trade-offs is generated. Furthermore, Monte Carlo simulation is introduced to conduct random scenario tests on each optimal scheme, ultimately transforming fuzzy risks into quantifiable indicators such as expected returns (NPV) and value at risk (VaR), considering probabilistic transformations of technological and market risks. Finally, through visualization and real-time feedback mechanisms, the decision results are mapped to geographic space and continuously compared with actual operating data. When the deviation exceeds a threshold, re-prediction and optimization are triggered, thus forming a dynamic closed loop that ensures the investment strategy always adapts to the actual development trajectory of the power grid.

[0050] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A power grid investment optimization method based on big data analysis, characterized in that: Includes the following steps: Data acquisition: Access multi-source heterogeneous data, perform edge preprocessing and cloud fusion and alignment, complete standardization processing, and store the data in a time-series database; Power grid condition prediction: Based on time series database, an LSTM model is built to predict future power load and new energy generation curves, thereby quantifying energy storage demand and evaluating project economics; Optimization and Risk Assessment: Based on the predicted power grid state, a set of optimal investment schemes is generated through a multi-objective optimization algorithm, and Monte Carlo simulation is used to quantitatively evaluate the risks and benefits of each scheme; Visualization and Feedback: The investment plan is displayed in geospatial visualization, and the power grid planning is optimized based on dynamic feedback from real-time monitoring data.

2. The power grid investment optimization method based on big data analysis according to claim 1, characterized in that: The data collection and acquisition includes the following steps: Multi-source data access: Systematically access various data sources, including internal power grid data, external environmental data, user and market data, and policy and planning data; Data preprocessing: The collected data is preprocessed by edge computing units deployed at the edge, including preliminary cleaning, aggregation and compression, and format conversion of high-frequency data; The cloud platform analyzes user and market data, as well as policy and planning data, and transforms them into processing data that correlates execution time with spatial relationships. Data fusion and storage: The data preprocessed at the edge is transmitted to the cloud platform and aligned with the data parsed by the cloud platform based on timestamps and geographic information. Then, normalization is used to standardize the data of different dimensions to a unified range. All processed data is stored in a time series database.

3. The power grid investment optimization method based on big data analysis according to claim 2, characterized in that: The internal power grid data includes historical and real-time load data, operating status, equipment ledgers, and fault records from SCADA and EMS systems; the external environment data includes solar radiation, wind speed, and temperature data accessed via API or sensor networks from meteorological bureaus / own monitoring points; the user and market data includes the number of electric vehicles, battery capacity, charging rate, and usage habits data from the charging pile operation platform, as well as electricity price information from the electricity market; and the policy and planning data includes energy policies, regulations, regional land space plans, and economic development indicators issued by governments at all levels.

4. The power grid investment optimization method based on big data analysis according to claim 1, characterized in that: The power grid state prediction includes the following steps: Dataset construction: Extract key features from the time series database, construct the model training dataset, and divide it into training set, validation set, and test set in chronological order; Data prediction: Using a long short-term memory network model, with the constructed dataset as input, the system predicts the load curves and renewable energy generation curves for the short and medium term. Indicator Calculation: Based on the predicted power generation and load, the difference between future time points is calculated to generate a supply-demand difference curve; by setting a power threshold, demand scheduling periods are identified; then, the energy storage scale is quantified based on the demand scheduling periods; finally, the NPV and IRR models are used to conduct a preliminary economic assessment of the proposed projects.

5. The power grid investment optimization method based on big data analysis according to claim 4, characterized in that: The dataset includes a load forecast dataset and a new energy power generation forecast dataset; The load forecast dataset includes time characteristics, meteorological characteristics, historical loads, and economic data; the new energy power generation forecast dataset includes historical power, wind power data, and photovoltaic data.

6. The power grid investment optimization method based on big data analysis according to claim 5, characterized in that: The identified demand scheduling period includes: When the supply-demand difference is greater than zero and greater than the power threshold, it is marked as a charging scheduling period; When the supply-demand difference is less than zero and its absolute value is greater than the power threshold, it is marked as a discharge scheduling period; When the absolute value of the supply-demand difference is less than or equal to the power threshold, it is marked as a balanced state.

7. The power grid investment optimization method based on big data analysis according to claim 6, characterized in that: The quantification of energy storage scale includes: Traverse the discharge scheduling period, find the value with the largest power shortage among all time points as the power demand value, classify the discharge scheduling period into different shortage event groups, and then calculate the sum of the power shortage values ​​of all time points in each shortage event group. Select the value with the largest sum of power shortage values ​​as the energy capacity demand value.

8. The power grid investment optimization method based on big data analysis according to claim 7, characterized in that: The preliminary economic assessment of the proposed project using NPV and IRR models includes: By constructing a cash flow model, calculating key economic indicators, including net present value and internal rate of return, and setting a minimum economic threshold, projects can be screened. In the cash flow model, the investment cost is estimated based on energy storage power and capacity requirements, with reference to the current unit cost.

9. The power grid investment optimization method based on big data analysis according to claim 1, characterized in that: The optimization and risk assessment include the following steps: Optimization problem definition: Clearly define decision variables, including the scale of the energy storage system, the lines that need to be upgraded, and the equipment selection; construct a multi-objective function, including minimizing the total investment cost, maximizing the return on investment, maximizing the power supply reliability, and minimizing the overall risk level; set constraints, including the budget ceiling and the physical energy storage capacity, where the constraint on the physical energy storage capacity is set as the total energy storage demand minus the virtual energy storage compensation parameter; Among them, the virtual energy storage compensation parameter is the alternative energy storage capacity calculated based on electric vehicle data; Multi-objective optimization solution: Genetic algorithm is used for multi-objective optimization solution. First, a population representing different investment schemes is randomly initialized. Then, the merits of each scheme are evaluated through the fitness function. By simulating the natural selection process, selection, crossover and mutation operations are performed on the population. Iterative evolution is carried out and finally converges to a set of Pareto optimal solutions. Global Risk Assessment: A global risk assessment is performed on the Pareto optimal solution set obtained through optimization. First, risk variables are identified, including technological risk, market risk, policy risk, and environmental and social risks. Then, based on the risk variables, the Monte Carlo simulation method is used to conduct random sampling simulations, calculate the probability distribution of returns for each investment option, and output the expected NPV and VaR values. Sort label: Read the expected NPV and VaR values ​​of each scheme, plot all schemes on a two-dimensional scatter plot, and then automatically classify the schemes falling into different regions into classification groups through a clustering algorithm, and add a strategy label field to each scheme; The classification groups include: high-risk, high-return, low-risk, stable, and balanced.

10. The power grid investment optimization method based on big data analysis according to claim 1, characterized in that: The visualization and feedback include the following steps: Geospatial visualization: Based on the Pareto optimal solution set, all investment plans are obtained, and the geometric data in the plans are converted into standard geographic information format and plotted on a GIS map; Real-time feedback and adjustment: After the investment plan is implemented, the actual operating data of the power grid is continuously monitored through IoT devices, and deviation thresholds are set for early warning. When the deviation threshold is reached, the latest actual data is used as input to re-execute the power grid state prediction.

Images