Electric power spot market electric power transaction intelligent electricity price prediction and electric power dispatching system and method

Abstract

The invention discloses an electric power spot market electric power transaction intelligent electricity price prediction and electric power dispatching system and method. The system comprises a data acquisition module, a preprocessing module, an electricity price prediction module, a load prediction module, a dispatching optimization module, a transaction decision module, a real-time monitoring module and a risk management module. The method comprises the following steps: step 1, multi-source data acquisition and standardization processing; 2, carrying out the collaborative prediction of the electricity price and the load; step 3, scheduling optimization and transaction strategy generation; 4, real-time monitoring and risk hedging feedback are carried out; according to the method, nonlinear and high-dimensional data can be effectively processed, the adaptability to distributed energy access and real-time meteorological changes is improved, new energy power generation data and market transaction strategies are effectively integrated, data processing can be fully carried out when complex fluctuation of an electric power spot market is coped, the model generalization ability is improved, and the method is suitable for large-scale popularization and application. Optimization of multiple time scales is achieved, and supplement of electricity price prediction and power dispatching is achieved.

Description

Technical Field

[0001] This invention belongs to the field of electricity price forecasting and power dispatching technology, and in particular relates to an intelligent electricity price forecasting and power dispatching system and method for electricity spot market electricity trading that integrates industrial data acquisition, processing and artificial intelligence algorithms. Background Technology

[0002] As is well known, in the electricity market, power dispatch optimization aims to achieve efficient operation of the power system, ensure a balance between power supply and demand and grid stability, while electricity price forecasting helps electricity market participants make scientific and rational decisions, improving the efficiency and competitiveness of the electricity market. Therefore, electricity price forecasting and power dispatch are two core functional modules supporting the safe and economical operation of modern power systems, and are important links in achieving stable supply, reasonable pricing, and optimized resource allocation.

[0003] Electricity market price forecasting involves analyzing historical price data, real-time load, and power transmission status to predict future price trends. This forecasting is crucial for power companies and energy traders as it helps them make more reliable decisions. Electricity price forecasting technology, through multi-model fusion algorithms, improves the accuracy of price predictions, enabling more precise forecasting of price fluctuations and providing power companies, energy traders, and large power-consuming enterprises with more reliable decision-making support. Power dispatching refers to the rational arrangement of power generation, transmission, and distribution plans based on the power system's operating status and demand to ensure the safe, stable, and economical operation of the power system. Its main responsibilities include: grid monitoring, fault handling, load forecasting, dispatching planning, emergency response, data recording and analysis, and communication and coordination. Through these functions, power dispatching ensures the balanced and stable operation of the power system, optimizes the utilization rate of power generation equipment, reduces operating costs, and promotes the consumption of renewable energy. Therefore, effective electricity price forecasting and dispatching not only reflect market expectations and anticipated market operating status but also promote the rational allocation of resources, the arrangement of effective power generation plans, and the formulation of reasonable pricing strategies by participants in the electricity market, which is of great significance for improving the efficiency of the electricity market.

[0004] Currently, traditional power trading forecasting and dispatching systems mostly rely on single machine learning models or statistical methods, resulting in insufficient forecast accuracy and slow dynamic response. Existing electricity price forecasting technologies, such as those based on time series analysis or support vector machines, struggle to handle nonlinear and high-dimensional data, leading to significant errors during peak periods. Load forecasting methods often employ similar day methods or artificial neural networks, but their adaptability to distributed energy integration and real-time weather changes is poor. Power dispatching systems generally adopt a hierarchical distributed architecture, but their level of intelligence is limited, failing to effectively integrate new energy generation data with market trading strategies, and fault diagnosis relies on preset rule bases, lacking autonomous learning capabilities. When dealing with the complex fluctuations of the electricity spot market, current systems often suffer from delayed dispatching decisions due to insufficient data processing and weak model generalization capabilities, failing to achieve optimization across multiple time scales. Furthermore, existing combined forecasting algorithms, such as the combination of Gaussian mixture models and lightweight gradient boosters, have not yet solved the problem of feature engineering redundancy and lack dynamic parameter adjustment mechanisms for power trading scenarios, making it difficult to meet the dual requirements of real-time trading and risk control.

[0005] Therefore, facing the problems of insufficient prediction accuracy and slow dynamic response, how to effectively process nonlinear and high-dimensional data, improve adaptability to distributed energy access and real-time weather changes, effectively integrate new energy generation data with market trading strategies, and fully process data to improve model generalization ability and achieve multi-timescale optimization when dealing with complex fluctuations in the electricity spot market, thus realizing the complementary goals of electricity price forecasting and power dispatching, has become a pressing problem for technical personnel in the field of electricity market price forecasting and power dispatching. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the technical problem this invention aims to solve is to provide an intelligent electricity price forecasting and power dispatching system and method for electricity spot market transactions. This system and method can effectively process nonlinear and high-dimensional data, improve adaptability to distributed energy access and real-time weather changes, effectively integrate new energy power generation data with market trading strategies, fully process data when dealing with complex fluctuations in the electricity spot market, improve model generalization ability, achieve multi-timescale optimization, and realize the complementary purpose of electricity price forecasting and power dispatching.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a smart electricity price prediction and power dispatch system for electricity spot market transactions, comprising:

[0008] The data acquisition module is used to collect meteorological data, unit output data, user load curves and historical transaction records in real time, and store dynamic data streams, converting unstructured data into a unified format;

[0009] The preprocessing module is used to eliminate time offsets in multi-source data, identify outliers, and remove outliers by setting the threshold to three times the standard deviation of the data distribution. It also performs dimensionless processing on the data to generate a standardized data stream and transmit it to the electricity price forecasting module and the load forecasting module.

[0010] The electricity price forecasting module is used to divide historical electricity price data into three clusters according to seasonal characteristics, fit a linear trend term with regularization parameters, and finally filter variables based on feature importance and output the electricity price forecasting results.

[0011] The load forecasting module is used to process nonlinear load characteristics and filter similar daily load curves based on dynamic time curvature distance. The forecasting results generated by the electricity price forecasting module and the load forecasting module are synchronously transmitted to the scheduling optimization module and the transaction decision module through a parallel interface.

[0012] The scheduling optimization module, based on the dynamic computer group output margin of the triangular membership function, uses an adaptive neural network unit with a structure where the number of hidden layer nodes is 1.2 times the number of input features to optimize the microgrid scheduling strategy;

[0013] The trading decision module is used to construct the Pareto frontier for spot and medium- to long-term trading. The risk value assessment unit simulates the maximum potential loss by backtracking through 30 days of historical data, generates scheduling instructions and trading strategies, and inputs them into the real-time monitoring module.

[0014] The real-time monitoring module is used to track the power grid's operating status and predict trends for the next 10 minutes by using forget gate parameters, and to detect abnormal states.

[0015] The risk management module is used to randomly sample electricity price fluctuations when the real-time monitoring module detects anomalies, dynamically calculate the optimal option position, and finally feed the risk hedging signal back to the scheduling optimization module to adjust the output parameters and update the weight coefficient of the trading decision module, forming a closed-loop control.

[0016] The aforementioned intelligent electricity price prediction and power dispatch system for electricity spot market transactions comprises a data acquisition module consisting of a multi-source data interface unit, a data caching unit, and a transmission protocol conversion unit. The multi-source data interface unit collects meteorological data, generator output data, user load curves, and historical transaction records in real time through the power Internet of Things protocol. The data caching unit uses a time window mechanism to store dynamic data streams. The transmission protocol conversion unit converts unstructured data into a unified format. The data throughput is designed to be 10,000 records per second, with a packet loss rate of less than 0.01%.

[0017] The aforementioned intelligent electricity price prediction and power dispatch system for the electricity spot market includes a preprocessing module comprising a time-series alignment unit, an anomaly detection unit, and a normalization unit. The time-series alignment unit uses a dynamic time warping algorithm to eliminate time offsets in multi-source data. The anomaly detection unit identifies outliers based on the isolated forest algorithm, with a threshold of three times the standard deviation of the data distribution. The normalization unit dedimensionalizes the data using the Z-Score standardization formula.

[0018]

[0019] Where x is the original feature value, μ is the feature mean, and σ is the standard deviation.

[0020] The aforementioned intelligent electricity price prediction and dispatch system for the electricity spot market comprises a price prediction module consisting of a Gaussian mixture clustering unit, a ridge regression preliminary prediction unit, and a light gradient booster optimization unit. The Gaussian mixture clustering unit divides historical electricity price data into three clusters based on seasonal characteristics, with the following probability density function:

[0021]

[0022] Where K=3 is the number of clusters, w k Let μ be the weight of the k-th cluster. k and Σ k These are the mean and covariance matrices, respectively.

[0023] The ridge regression preliminary prediction unit uses a regularization parameter λ = 0.5 to fit a linear trend term, and the objective function is:

[0024] min(||y-Xβ|| 2 +λ||β|| 2 );

[0025] Where y is the actual electricity price, X is the characteristic matrix, and β is the regression coefficient;

[0026] The Light Gradient Boosting Machine (LGBM) optimization unit selects input variables based on feature importance. Its loss function has a leaf node number penalty coefficient γ = 0.1, a weight regularization coefficient λ = 0.01, and a tree depth limit of 6 layers.

[0027] The aforementioned intelligent electricity price forecasting and power dispatching system for the electricity spot market integrates a support vector regression unit and a similar day matching unit in its load forecasting module. The support vector regression unit uses a radial basis function kernel to handle nonlinear load characteristics, and the kernel function expression is as follows:

[0028]

[0029] Where σ = 0.8 is the kernel function bandwidth, x i ,xj (For the sample feature vector), the penalty factor C = 1.5 is used to balance model complexity and error tolerance; the similar day matching unit filters historical similar days based on dynamic time bending distance (DTW), and its distance threshold is set to 15% of the total length of the load curve.

[0030] The aforementioned intelligent electricity price prediction and power dispatch system for the electricity spot market includes a dispatch optimization module comprising a fuzzy control unit and an adaptive neural network unit. The fuzzy control unit defines a triangular membership function to describe the unit output margin, with the function form as follows:

[0031]

[0032] Where x is the input variable, a is the membership center, and b is the support set width, which is dynamically adjusted according to the unit capacity;

[0033] The adaptive neural network unit adopts a dual hidden layer structure, with the number of hidden layer nodes being 1.2 times the number of input features, and the activation function being Leaky ReLU.

[0034]

[0035] The adaptive adjustment range of the learning rate is 0.001 to 0.01.

[0036] The aforementioned intelligent electricity price prediction and power dispatch system for the electricity spot market comprises a trading decision module consisting of a multi-objective optimization unit and a risk value assessment unit. The multi-objective optimization unit constructs the Pareto front of the spot and medium-to-long-term trading portfolios, with the objective function being:

[0037] min(w1C spot +w2VaR);

[0038] Where w1 = 0.7 is the spot transaction cost weight, w2 = 0.3 is the value at risk weight, and the VaR confidence level is 95%;

[0039] The risk value assessment unit uses historical simulation to estimate the maximum potential loss, with a backtesting window of 30 trading days.

[0040] The aforementioned intelligent electricity price prediction and power dispatch system for the electricity spot market includes a real-time monitoring module comprising an edge computing unit and a Long Short-Term Memory (LSTM) early warning unit. The edge computing unit is deployed at the substation side, supporting millisecond-level data refresh and local decision-making. The LSTM early warning unit has 64 hidden layer neurons, and the forget gate parameter expression is as follows:

[0041] f t =σ(W f ·[h t-1 ,xt ]+b f );

[0042] Among them, W f Let b be the weight matrix. f σ is the bias term, σ is the Sigmoid function, the initialization uses the Xavier normal distribution, and the time series prediction step size is 10 minutes.

[0043] The aforementioned intelligent electricity price prediction and dispatch system for the electricity spot market employs a Monte Carlo simulation unit and a hedging strategy generation unit in its risk management module. The Monte Carlo simulation unit performs 10,000 random samplings of electricity price volatility, and its Brownian motion discretization formula is as follows:

[0044]

[0045] Among them, S t Given the current electricity price, μ is the drift rate, σ is the volatility, ∈ ~N(0,1), and the discretization step size Δt = 1 hour;

[0046] The hedging strategy generation unit calculates the optimal option position using a dynamic programming algorithm, with the objective function being:

[0047] min(0.75C hedge +0.25R exposure );

[0048] Among them, C hedge To hedge costs, R exposure (remaining risk exposure).

[0049] A method for intelligent electricity price forecasting and power dispatching in the electricity spot market includes the following steps:

[0050] Step 1: Multi-source data acquisition and standardization processing;

[0051] Step 2: Co-forwarding of electricity prices and load;

[0052] Step 3: Scheduling optimization and trading strategy generation;

[0053] Step 4: Real-time monitoring and risk hedging feedback;

[0054] In step one, the data acquisition module acquires meteorological data, unit output data and user load curves in real time through the power Internet of Things protocol. After the transmission protocol conversion unit unifies the format, the data is stored in the cache unit. The preprocessing module simultaneously starts the timing alignment unit to eliminate the time offset of multi-source data. The anomaly detection unit removes outliers based on a three-standard-deviation threshold. The normalization processing unit performs dimensionless processing on the data using the Z-Score standardization formula, generates a standardized data stream and transmits it to the electricity price prediction module and the load prediction module.

[0055] In step two, the Gaussian mixture clustering unit of the electricity price prediction module divides the historical electricity price data into three clusters according to seasonal characteristics, the ridge regression unit fits the linear trend term with a regularization parameter λ = 0.5, and the light gradient booster optimization unit selects variables based on feature importance and outputs the electricity price prediction results; the load prediction module uses the support vector regression unit to process nonlinear features with a kernel function bandwidth σ = 0.8, and the similar day matching unit selects similar day load curves based on dynamic time curvature distance. The prediction results generated by the two modules are synchronously transmitted to the scheduling optimization module and the transaction decision module through a parallel interface.

[0056] In step three, the fuzzy control unit of the scheduling optimization module is based on the dynamic computer group output margin of the triangular membership function, and the adaptive neural network unit optimizes the microgrid scheduling strategy with a structure in which the number of hidden layer nodes is 1.2 times the number of input features; the multi-objective optimization unit of the trading decision module constructs the Pareto frontier of spot and medium- and long-term trading, and the risk value assessment unit simulates the maximum potential loss through 30 days of historical data backtracking, generates scheduling instructions and trading strategies and inputs them into the real-time monitoring module;

[0057] In step four, the edge computing unit of the real-time monitoring module deploys a long short-term memory network early warning unit, which tracks the power grid operation status and predicts the trend in the next 10 minutes through forget gate parameters. When an anomaly is detected, the Monte Carlo simulation unit of the risk management module is triggered to randomly sample the electricity price fluctuation. The hedging strategy generation unit dynamically calculates the optimal option position, and finally feeds the risk hedging signal back to the scheduling optimization module to adjust the output parameters and update the weight coefficient of the trading decision module, forming a closed-loop control.

[0058] The advantages of the intelligent electricity price prediction and power dispatch system and method for electricity spot market trading of this invention are as follows: The prediction module of this invention uses a combination of prediction algorithms such as Gaussian mixture clustering unit, ridge regression unit, and light gradient booster optimization unit to accurately predict electricity prices and loads. The electricity price prediction can achieve smooth fitting during normal periods and convergence of errors during peak periods. The load prediction uses support vector regression unit and similar day matching unit to handle nonlinear features and screen historical load curves. These prediction results provide key data support for the dispatch optimization module, enabling the dispatch optimization module to dynamically adjust the triangular membership function parameters through the fuzzy control unit based on accurate prediction information, and combine it with a dual hidden layer adaptive neural network to optimize the unit output strategy, thereby achieving flexible unit dispatch and microgrid mode switching, and improving energy synergy efficiency.

[0059] Simultaneously, the operational status of the scheduling optimization module and the results of coordinated energy scheduling are fed back to the prediction module, further helping it calibrate and optimize its prediction model, making its predictions more closely aligned with actual grid operation and scheduling conditions. For example, the LSTM early warning unit deployed in the real-time monitoring module achieves millisecond-level data refresh and 10-minute time-series prediction, while the edge computing unit dynamically tracks grid state deviations through forget gate parameters. This real-time monitoring data provides the prediction module with real-time grid state information, enabling timely adjustments to its predictions, and also provides a basis for the real-time decision-making of the scheduling optimization module. The adjustments made by the scheduling optimization module based on this data, in turn, affect subsequent predictions. Compared to traditional technologies, this invention possesses stronger data mining capabilities, faster dynamic response speeds, and a more comprehensive global coordination mechanism, providing end-to-end intelligent support for power trading. Attached Figure Description

[0060] Figure 1 This is a schematic diagram illustrating the interaction principle between the various components of the intelligent prediction and scheduling system for power trading of the present invention.

[0061] Figure 2 This is a schematic diagram illustrating the operating principles of each component unit in the data acquisition module 1 and the preprocessing module 2.

[0062] Figure 3 The diagram shows the operating principle of each component unit in the electricity price forecasting module 3 and the load forecasting module 4.

[0063] Figure 4 This is a schematic diagram illustrating the operational principles of each component unit in the scheduling optimization module 5 and the transaction decision module 6.

[0064] Figure 5 This is a schematic diagram illustrating the operating principles of each component unit in the real-time monitoring module 7 and the risk management module 8.

[0065] Figure 6 This diagram illustrates the main data interaction in the fully automated processing of the intelligent electricity price forecasting and power dispatching system of this invention. Detailed Implementation

[0066] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] In this invention, unless otherwise stated, directional terms such as "upper" and "lower" generally refer to the upper and lower positions of the device in its actual use or operating state, specifically the drawing directions in the accompanying drawings; while "inner" and "outer" refer to the outline of the device. Furthermore, in the description of this application, the term "comprising" means "including but not limited to". The terms first, second, third, etc., are used merely as illustrative purposes and do not impose numerical requirements or establish an order. The term "multiple" means "two or more".

[0068] As is well known, there is a close relationship between electricity market price forecasting and power dispatching. First, price forecasting has a significant impact on power dispatching. The accuracy of price forecasting directly affects the supply and demand balance of the electricity market and power dispatching decisions. Through price forecasting, a better understanding of market supply and demand can be achieved, thereby adjusting generation plans and load allocation to ensure the stable operation of the power system. For example, accurate price forecasting can help dispatchers better arrange power plant operation plans and avoid economic losses caused by price fluctuations. Second, power dispatching, in turn, also affects the accuracy of price forecasting. Power dispatching decisions directly affect the supply and demand situation of the electricity market, and thus affect electricity prices. For example, grid congestion and changes in unit composition can lead to price fluctuations, and these factors need to be fully considered in price forecasting. If power dispatching cannot respond to these changes in a timely manner, it may lead to inaccurate price forecasts, thereby affecting the decisions of market participants. Furthermore, the goals of both price forecasting and power dispatching are to ensure the stable operation of the power system and maximize economic benefits. Price forecasting, by analyzing historical data and market trends, helps dispatchers make more rational decisions; while power dispatching, through actual operation, ensures the safety and stability of the power grid. These two aspects complement each other and jointly promote the healthy development of the electricity market. Based on this, the present invention utilizes industrial data acquisition, processing, and the integration of artificial intelligence algorithms to achieve the aforementioned technical objectives through an intelligent electricity price prediction and power dispatch system and method for the electricity spot market.

[0069] like Figure 1-5 As shown, a smart electricity price forecasting and power dispatching system for electricity spot market trading includes:

[0070] The data acquisition module is used to collect meteorological data, unit output data, user load curves and historical transaction records in real time, and store dynamic data streams, converting unstructured data into a unified format;

[0071] The preprocessing module is used to eliminate time offsets in multi-source data, identify outliers, and remove outliers by setting the threshold to three times the standard deviation of the data distribution. It also performs dimensionless processing on the data to generate a standardized data stream and transmit it to the electricity price forecasting module and the load forecasting module.

[0072] The electricity price forecasting module is used to divide historical electricity price data into three clusters according to seasonal characteristics, fit a linear trend term with regularization parameters, and finally filter variables based on feature importance and output the electricity price forecasting results.

[0073] The load forecasting module is used to process nonlinear load characteristics and filter similar daily load curves based on dynamic time curvature distance. The forecasting results generated by the electricity price forecasting module and the load forecasting module are synchronously transmitted to the scheduling optimization module and the transaction decision module through a parallel interface.

[0074] The scheduling optimization module, based on the dynamic computer group output margin of the triangular membership function, uses an adaptive neural network unit with a structure where the number of hidden layer nodes is 1.2 times the number of input features to optimize the microgrid scheduling strategy;

[0075] The trading decision module is used to construct the Pareto frontier for spot and medium- to long-term trading. The risk value assessment unit simulates the maximum potential loss by backtracking through 30 days of historical data, generates scheduling instructions and trading strategies, and inputs them into the real-time monitoring module.

[0076] The real-time monitoring module is used to track the power grid's operating status and predict trends for the next 10 minutes by using forget gate parameters, and to detect abnormal states.

[0077] The risk management module is used to randomly sample electricity price fluctuations when the real-time monitoring module detects anomalies, dynamically calculate the optimal option position, and finally feed the risk hedging signal back to the scheduling optimization module to adjust the output parameters and update the weight coefficient of the trading decision module, forming a closed-loop control.

[0078] Specifically, the data acquisition module establishes a data connection with the preprocessing module, the preprocessing module establishes a data connection with the electricity price forecasting module and the load forecasting module respectively, the electricity price forecasting module and the load forecasting module establish a data connection with the dispatch optimization module and the transaction decision module respectively, the dispatch optimization module and the transaction decision module both establish a data connection with the real-time monitoring module, the real-time monitoring module establishes a data connection with the risk management module, and the risk management module establishes a data connection with the dispatch optimization module and the transaction decision module.

[0079] The data acquisition module consists of a multi-source data interface unit, a data caching unit, and a transmission protocol conversion unit. The multi-source data interface unit collects meteorological data, unit output data, user load curves, and historical transaction records in real time through power Internet of Things protocols (such as IEC 61850). The data caching unit uses a time window mechanism to store dynamic data streams. The transmission protocol conversion unit converts unstructured data into a unified format. The data throughput is designed to be 10,000 records per second, with a packet loss rate of less than 0.01%.

[0080] The preprocessing module includes a time alignment unit, an anomaly detection unit, and a normalization unit. The time alignment unit uses a dynamic time warping algorithm to eliminate time offsets in multi-source data. The anomaly detection unit identifies outliers based on the isolated forest algorithm, with a threshold of three standard deviations of the data distribution. The normalization unit dedimensionalizes the data using the Z-Score normalization formula.

[0081]

[0082] Where x is the original feature value, μ is the feature mean, and σ is the standard deviation.

[0083] The electricity price forecasting module consists of a Gaussian mixture clustering unit, a ridge regression preliminary forecasting unit, and a light gradient boosting machine optimization unit. The Gaussian mixture clustering unit divides historical electricity price data into three clusters based on seasonal characteristics, and its probability density function is:

[0084]

[0085] Where K=3 is the number of clusters, w k Let μ be the weight of the k-th cluster. k and Σ k These are the mean and covariance matrices, respectively.

[0086] The ridge regression preliminary prediction unit uses a regularization parameter λ = 0.5 to fit a linear trend term, and the objective function is:

[0087] min(||y-Xβ|| 2 +λ||β|| 2 );

[0088] Where y is the actual electricity price, X is the characteristic matrix, and β is the regression coefficient;

[0089] The Light Gradient Boosting Machine (LGBM) optimization unit selects input variables based on feature importance. Its loss function has a leaf node number penalty coefficient γ = 0.1, a weight regularization coefficient λ = 0.01, and a tree depth limit of 6 layers.

[0090] The load forecasting module integrates a support vector regression unit and a similar day matching unit. The support vector regression unit uses a radial basis function kernel to handle nonlinear load characteristics. The kernel function expression is as follows:

[0091]

[0092] Where σ = 0.8 is the kernel function bandwidth, x i ,x j (For the sample feature vector), the penalty factor C = 1.5 is used to balance model complexity and error tolerance; the similar day matching unit filters historical similar days based on dynamic time bending distance (DTW), and its distance threshold is set to 15% of the total length of the load curve.

[0093] The scheduling optimization module includes a fuzzy control unit and an adaptive neural network unit. The fuzzy control unit defines a triangular membership function to describe the unit's output margin, and the function form is as follows:

[0094]

[0095] Where x is the input variable, a is the membership center, and b is the support set width, which is dynamically adjusted according to the unit capacity;

[0096] The adaptive neural network unit adopts a dual hidden layer structure, with the number of hidden layer nodes being 1.2 times the number of input features, and the activation function being Leaky ReLU.

[0097]

[0098] The adaptive adjustment range of the learning rate is 0.001 to 0.01.

[0099] The trading decision module consists of a multi-objective optimization unit and a risk value assessment unit. The multi-objective optimization unit constructs the Pareto front for spot and medium-to-long-term trading portfolios, with the objective function being:

[0100] min(w1C spot +w2VaR);

[0101] Where w1 = 0.7 is the spot transaction cost weight, w2 = 0.3 is the value at risk weight, and the VaR confidence level is 95%;

[0102] The risk value assessment unit uses historical simulation to estimate the maximum potential loss, with a backtesting window of 30 trading days.

[0103] The real-time monitoring module includes an edge computing unit and a Long Short-Term Memory (LSTM) early warning unit. The edge computing unit is deployed on the substation side, supporting millisecond-level data refresh and local decision-making. The LSTM early warning unit has 64 hidden layer neurons, and the forget gate parameter expression is:

[0104] f t =σ(W f ·[h t-1 x t ]+b f );

[0105] Among them, W f Let b be the weight matrix. f σ is the bias term, σ is the Sigmoid function, the initialization uses the Xavier normal distribution, and the time series prediction step size is 10 minutes.

[0106] The risk management module employs a Monte Carlo simulation unit and a hedging strategy generation unit. The Monte Carlo simulation unit performs 10,000 random samples of electricity price volatility, and its Brownian motion discretization formula is as follows:

[0107]

[0108] Among them, S t Given the current electricity price, μ is the drift rate, σ is the volatility, ∈ ~N(0,1), and the discretization step size Δt = 1 hour;

[0109] The hedging strategy generation unit calculates the optimal option position using a dynamic programming algorithm, with the objective function being:

[0110] min(0.75C hedge +0.25R exposure );

[0111] Among them, C hedge To hedge costs, R exposure (remaining risk exposure).

[0112] like Figure 6 As shown, a method for intelligent electricity price forecasting and power dispatching in the electricity spot market includes the following steps:

[0113] Step 1: Multi-source data acquisition and standardization processing;

[0114] Step 2: Co-forwarding of electricity prices and load;

[0115] Step 3: Scheduling optimization and trading strategy generation;

[0116] Step 4: Real-time monitoring and risk hedging feedback;

[0117] In step one, the data acquisition module acquires meteorological data, unit output data and user load curves in real time through the power Internet of Things protocol. After being formatted by the transmission protocol conversion unit, the data is stored in the cache unit. The preprocessing module simultaneously starts the timing alignment unit to eliminate the time offset of multi-source data. The anomaly detection unit removes outliers based on a three-standard-deviation threshold. The normalization processing unit performs dimensionless processing on the data using the Z-Score standardization formula, generates a standardized data stream, and transmits it to the electricity price forecasting module and the load forecasting module.

[0118] In step two, the Gaussian mixture clustering unit of the electricity price forecasting module divides the historical electricity price data into three clusters according to seasonal characteristics. The ridge regression unit fits the linear trend term with a regularization parameter λ = 0.5. The light gradient booster optimization unit selects variables based on feature importance and outputs the electricity price forecasting results. The load forecasting module uses the support vector regression unit to process nonlinear features with a kernel function bandwidth σ = 0.8. The similar day matching unit selects similar day load curves based on dynamic time curvature distance. The forecasting results generated by the two modules are synchronously transmitted to the scheduling optimization module and the transaction decision module through a parallel interface.

[0119] In step three, the fuzzy control unit of the scheduling optimization module dynamically calculates the power margin of the computer group based on the triangular membership function, and the adaptive neural network unit optimizes the microgrid scheduling strategy with a structure in which the number of hidden layer nodes is 1.2 times the number of input features. The multi-objective optimization unit of the trading decision module constructs the Pareto front for spot and medium-to-long-term trading, and the risk value assessment unit simulates the maximum potential loss through 30 days of historical data backtracking, generates scheduling instructions and trading strategies, and inputs them into the real-time monitoring module.

[0120] In step four, the edge computing unit of the real-time monitoring module deploys a long short-term memory network early warning unit, which tracks the power grid operation status and predicts the trend in the next 10 minutes through forget gate parameters. When an anomaly is detected, the Monte Carlo simulation unit of the risk management module is triggered to randomly sample the electricity price fluctuation. The hedging strategy generation unit dynamically calculates the optimal option position, and finally feeds the risk hedging signal back to the scheduling optimization module to adjust the output parameters and update the weight coefficient of the trading decision module, forming a closed-loop control.

[0121] This invention's data acquisition module is used to collect meteorological data, generator output data, and user load curves in real time via a power Internet of Things (IoT) protocol. It includes a multi-source data interface unit, a data caching unit, and a transmission protocol conversion unit. The multi-source data interface unit collects heterogeneous data from multiple sources. The data caching unit uses a time window mechanism to store dynamic data streams. The transmission protocol conversion unit unifies the data format and ensures a high throughput of tens of thousands of records per second. The preprocessing module eliminates data time skew, removes outliers, and performs standardization. It includes a time-series alignment unit, an anomaly detection unit, and a normalization unit. The time-series alignment unit aligns time-series data using a dynamic time warping algorithm. The anomaly detection unit identifies outliers based on a three-standard-deviation threshold. The normalization unit eliminates dimensional differences using the Z-Score formula.

[0122] The electricity price forecasting module predicts electricity price fluctuation trends and includes a Gaussian mixture clustering unit, a ridge regression preliminary forecasting unit, and a light gradient boosting machine optimization unit. The Gaussian mixture clustering unit divides historical electricity price data into clusters based on seasonal characteristics, the ridge regression preliminary forecasting unit fits a linear trend using regularization parameters, and the light gradient boosting machine optimization unit improves forecast accuracy through feature importance filtering. The load forecasting module generates load curves and includes a support vector regression unit and a similar day matching unit. The support vector regression unit uses a radial basis function kernel to handle nonlinear features, and the similar day matching unit filters historical similar days based on dynamic time curvature distance.

[0123] The scheduling optimization module dynamically adjusts the unit's output strategy and includes a fuzzy control unit and an adaptive neural network unit. The fuzzy control unit calculates the unit's output margin using a triangular membership function, while the adaptive neural network unit optimizes scheduling parameters using a dual-hidden-layer structure. The trading decision module balances trading costs and risks and includes a multi-objective optimization unit and a risk value assessment unit. The multi-objective optimization unit constructs the Pareto front for spot and medium-to-long-term trading, while the risk value assessment unit evaluates the maximum potential loss using historical simulation methods.

[0124] The real-time monitoring module is used for real-time tracking of the power grid status. It includes an edge computing unit and an LSTM early warning unit. The edge computing unit processes data locally, while the LSTM early warning unit predicts the power grid trend for the next 10 minutes using forget gate parameters. The risk management module is used to generate risk hedging strategies. It includes a Monte Carlo simulation unit and a hedging strategy generation unit. The Monte Carlo simulation unit performs tens of thousands of random samplings of electricity price fluctuations, while the hedging strategy generation unit optimizes option positions using a dynamic programming algorithm to reduce risk exposure.

[0125] The advantages of this invention are: it achieves intelligent closed-loop management of the entire power trading process through multi-module collaboration. The data acquisition module relies on the power Internet of Things protocol to collect meteorological data, generator output data, and user load curves in real time. It integrates heterogeneous data sources such as power market prices, policy documents, and transaction contracts through a multi-source data interface unit, and achieves high-throughput transmission of tens of thousands of records per second through a transmission protocol conversion unit. The preprocessing module uses a dynamic time warping algorithm to align time-series data, uses an isolated forest algorithm with a threshold of three standard deviations to remove outliers, and eliminates dimensional differences through Z-Score standardization. The electricity price forecasting module uses a Gaussian mixture clustering unit to divide electricity price data clusters according to seasonal characteristics, a ridge regression unit to fit linear trend terms with a regularization parameter λ = 0.5, and a LightGBM optimization unit to improve prediction accuracy through feature importance screening. The load forecasting module uses a support vector regression unit combined with a radial basis function kernel to handle nonlinear features, and a similar day matching unit to screen historical load curves based on a DTW distance threshold.

[0126] The scheduling optimization module dynamically adjusts the triangular membership function parameters through a fuzzy control unit, combining this with a dual-hidden-layer adaptive neural network to optimize unit output strategies. The trading decision module constructs a Pareto front for spot and medium-to-long-term trading, employs historical simulation to assess risk value, and generates tens of thousands of random sampling paths through Monte Carlo simulation. The real-time monitoring module deploys an LSTM early warning unit to achieve millisecond-level data updates and 10-minute time-series predictions, while the edge computing unit dynamically tracks grid state deviations through forget gate parameters. The dynamic risk exposure assessment results are fed back to the hedging strategy generation unit, forming a closed-loop control system from data acquisition to risk hedging.

[0127] The settlement strategy selection module constructs a state space and reward function based on a reinforcement learning algorithm, and intelligently switches between contract settlement, spot settlement, and hybrid modes through a Q-learning algorithm. The settlement price calculation unit integrates dynamic adjustment factors to weight multiple electricity prices, and the settlement fee calculation unit integrates transmission fees and ancillary service fees to generate the final bill. The settlement report module automatically generates electronic transaction vouchers through blockchain smart contracts, and combines a Xavier normal distribution initialization algorithm to ensure data transmission security, ultimately achieving fully automated processing from load forecasting to transaction settlement.

[0128] In summary, this invention effectively overcomes the limitations of traditional single-model prediction through a combined prediction algorithm, achieving convergence of peak-period errors in electricity price forecasting while enhancing the adaptability of load forecasting to fluctuations in renewable energy and sudden weather changes. The collaborative scheduling strategy of fuzzy control and adaptive neural networks shortens unit response time, supports flexible switching of microgrid operating modes, and improves energy synergy efficiency. The real-time monitoring system breaks through the static constraints of traditional rule bases, capturing abnormal signs in the power grid and initiating protection procedures in advance through an autonomous learning mechanism. The trading decision module innovatively combines risk hedging with multi-objective optimization, systematically reducing market risk exposure while controlling transaction costs. Compared to traditional technologies, this invention possesses stronger data mining capabilities, faster dynamic response speed, and a more comprehensive global collaborative mechanism, providing intelligent support for the entire power trading chain.

[0129] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should be protected by the present invention.

Claims

1. A smart electricity price forecasting and power dispatching system for electricity spot market transactions, characterized in that, include: The data acquisition module is used to collect meteorological data, unit output data, user load curves and historical transaction records in real time, and store dynamic data streams, converting unstructured data into a unified format; The preprocessing module is used to eliminate time offsets in multi-source data, identify outliers, and remove outliers by setting the threshold to three times the standard deviation of the data distribution. It also performs dimensionless processing on the data to generate a standardized data stream and transmit it to the electricity price forecasting module and the load forecasting module. The electricity price forecasting module is used to divide historical electricity price data into three clusters according to seasonal characteristics, fit a linear trend term with regularization parameters, and finally filter variables based on feature importance and output the electricity price forecasting results. The load forecasting module is used to process nonlinear load characteristics and filter similar daily load curves based on dynamic time curvature distance. The forecasting results generated by the electricity price forecasting module and the load forecasting module are synchronously transmitted to the scheduling optimization module and the transaction decision module through a parallel interface. The scheduling optimization module, based on the dynamic computer group output margin of the triangular membership function, uses an adaptive neural network unit with a structure where the number of hidden layer nodes is 1.2 times the number of input features to optimize the microgrid scheduling strategy; The trading decision module is used to construct the Pareto frontier for spot and medium- to long-term trading. The risk value assessment unit simulates the maximum potential loss by backtracking through 30 days of historical data, generates scheduling instructions and trading strategies, and inputs them into the real-time monitoring module. The real-time monitoring module is used to track the power grid's operating status and predict trends for the next 10 minutes by using forget gate parameters, and to detect abnormal states. The risk management module is used to randomly sample electricity price fluctuations when the real-time monitoring module detects anomalies, dynamically calculate the optimal option position, and finally feed the risk hedging signal back to the scheduling optimization module to adjust the output parameters and update the weight coefficient of the trading decision module, forming a closed-loop control.

2. The intelligent electricity price forecasting and power dispatching system for the electricity spot market as described in claim 1, characterized in that: The data acquisition module consists of a multi-source data interface unit, a data caching unit, and a transmission protocol conversion unit. The multi-source data interface unit collects meteorological data, unit output data, user load curves, and historical transaction records in real time through the power Internet of Things protocol. The data caching unit uses a time window mechanism to store dynamic data streams. The transmission protocol conversion unit converts unstructured data into a unified format. The data throughput is designed to be 10,000 records per second, with a packet loss rate of less than 0.01%.

3. The intelligent electricity price forecasting and power dispatching system for the electricity spot market as described in claim 1, characterized in that: The preprocessing module includes a time alignment unit, an anomaly detection unit, and a normalization unit. The time alignment unit uses a dynamic time warping algorithm to eliminate time offsets in multi-source data. The anomaly detection unit identifies outliers based on the isolated forest algorithm, with a threshold of three times the standard deviation of the data distribution. The normalization unit dedimensionalizes the data using the Z-Score normalization formula. Where x is the original feature value, μ is the feature mean, and σ is the standard deviation.

4. The intelligent electricity price forecasting and power dispatching system for the electricity spot market as described in claim 1, characterized in that: The electricity price prediction module consists of a Gaussian mixture clustering unit, a ridge regression preliminary prediction unit, and a light gradient booster optimization unit. The Gaussian mixture clustering unit divides historical electricity price data into three clusters according to seasonal characteristics, and its probability density function is: Where K=3 is the number of clusters, w k μ is the weight of the k-th cluster. k and Σ k These are the mean and covariance matrices, respectively. The ridge regression preliminary prediction unit uses a regularization parameter λ = 0.5 to fit a linear trend term, and the objective function is: min(||y-Xβ|| 2 +λ||β|| 2 ); Where y is the actual electricity price, X is the characteristic matrix, and β is the regression coefficient; The Light Gradient Boosting Machine (LGBM) optimization unit selects input variables based on feature importance. Its loss function has a leaf node number penalty coefficient γ = 0.1, a weight regularization coefficient λ = 0.01, and a tree depth limit of 6 layers.

5. The intelligent electricity price forecasting and power dispatching system for electricity spot market trading according to claim 1, characterized in that: The load forecasting module integrates a support vector regression unit and a similar day matching unit. The support vector regression unit uses a radial basis function kernel to handle nonlinear load characteristics. The kernel function expression is as follows: Where σ = 0.8 is the kernel function bandwidth, x i ,x j (For the sample feature vector), the penalty factor C = 1.5 is used to balance model complexity and error tolerance; the similar day matching unit filters historical similar days based on dynamic time bending distance (DTW), and its distance threshold is set to 15% of the total length of the load curve.

6. The intelligent electricity price forecasting and power dispatching system for electricity spot market trading according to claim 1, characterized in that: The scheduling optimization module includes a fuzzy control unit and an adaptive neural network unit. The fuzzy control unit defines a triangular membership function to describe the unit's output margin, and the function form is as follows: Where x is the input variable, a is the membership center, and b is the support set width, which is dynamically adjusted according to the unit capacity; The adaptive neural network unit adopts a dual hidden layer structure, with the number of hidden layer nodes being 1.2 times the number of input features, and the activation function being Leaky ReLU. The adaptive adjustment range of the learning rate is 0.001 to 0.

01.

7. The intelligent electricity price forecasting and power dispatching system for electricity spot market trading according to claim 1, characterized in that: The trading decision module consists of a multi-objective optimization unit and a risk value assessment unit. The multi-objective optimization unit constructs the Pareto frontier of the spot and medium-to-long-term trading portfolios, with the objective function being: min(w1C spot +w2VaR); Where w1 = 0.7 is the spot transaction cost weight, w2 = 0.3 is the value at risk weight, and the VaR confidence level is 95%; The risk value assessment unit uses historical simulation to estimate the maximum potential loss, with a backtesting window of 30 trading days.

8. The intelligent electricity price forecasting and power dispatching system for electricity spot market trading according to claim 1, characterized in that: The real-time monitoring module includes an edge computing unit and a Long Short-Term Memory (LSTM) early warning unit. The edge computing unit is deployed on the substation side, supporting millisecond-level data refresh and local decision-making. The LSTM early warning unit has 64 hidden layer neurons, and the forget gate parameter expression is: f t =σ(W f ·[h t-1 ,x t ]+b f ); Among them, W f Let b be the weight matrix. f σ is the bias term, σ is the Sigmoid function, the initialization uses the Xavier normal distribution, and the time series prediction step size is 10 minutes.

9. The intelligent electricity price forecasting and power dispatching system for electricity spot market trading according to claim 1, characterized in that: The risk management module employs a Monte Carlo simulation unit and a hedging strategy generation unit. The Monte Carlo simulation unit performs 10,000 random samples of electricity price volatility, and its Brownian motion discretization formula is as follows: Among them, S t Given the current electricity price, μ is the drift rate, σ is the volatility, ∈ ~N(0,1), and the discretization step size Δt = 1 hour; The hedging strategy generation unit calculates the optimal option position using a dynamic programming algorithm, with the objective function being: min(0.75C hedge +0.25R exposure ); Among them, C hedge To hedge costs, R exposure (remaining risk exposure).

10. A method for electricity price forecasting and power dispatching using the system according to any one of claims 1-9, characterized in that, Includes the following steps: Step 1: Multi-source data acquisition and standardization processing; Step 2: Co-forwarding of electricity prices and load; Step 3: Scheduling optimization and trading strategy generation; Step 4: Real-time monitoring and risk hedging feedback; In step one, the data acquisition module acquires meteorological data, unit output data and user load curves in real time through the power Internet of Things protocol. After the transmission protocol conversion unit unifies the format, the data is stored in the cache unit. The preprocessing module simultaneously starts the timing alignment unit to eliminate the time offset of multi-source data. The anomaly detection unit removes outliers based on a three-standard-deviation threshold. The normalization processing unit performs dimensionless processing on the data using the Z-Score standardization formula, generates a standardized data stream and transmits it to the electricity price prediction module and the load prediction module. In step two, the Gaussian mixture clustering unit of the electricity price prediction module divides the historical electricity price data into three clusters according to seasonal characteristics, the ridge regression unit fits the linear trend term with a regularization parameter λ = 0.5, and the light gradient booster optimization unit selects variables based on feature importance and outputs the electricity price prediction results; the load prediction module uses the support vector regression unit to process nonlinear features with a kernel function bandwidth σ = 0.8, and the similar day matching unit selects similar day load curves based on dynamic time curvature distance. The prediction results generated by the two modules are synchronously transmitted to the scheduling optimization module and the transaction decision module through a parallel interface. In step three, the fuzzy control unit of the scheduling optimization module is based on the dynamic computer group output margin of the triangular membership function, and the adaptive neural network unit optimizes the microgrid scheduling strategy with a structure in which the number of hidden layer nodes is 1.2 times the number of input features; the multi-objective optimization unit of the trading decision module constructs the Pareto frontier of spot and medium- and long-term trading, and the risk value assessment unit simulates the maximum potential loss through 30 days of historical data backtracking, generates scheduling instructions and trading strategies and inputs them into the real-time monitoring module; In step four, the edge computing unit of the real-time monitoring module deploys a long short-term memory network early warning unit, which tracks the power grid operation status and predicts the trend in the next 10 minutes through forget gate parameters. When an anomaly is detected, the Monte Carlo simulation unit of the risk management module is triggered to randomly sample the electricity price fluctuation. The hedging strategy generation unit dynamically calculates the optimal option position, and finally feeds the risk hedging signal back to the scheduling optimization module to adjust the output parameters and update the weight coefficient of the trading decision module, forming a closed-loop control.

Images