New energy power consumption and energy storage adjustment prediction method based on machine learning

Abstract

The invention discloses a new energy power consumption and energy storage adjustment prediction method based on machine learning, which belongs to the technical field of power, and comprises four levels of data acquisition, processing, analysis and visualization, the data of a power grid and an energy storage system are acquired, an active power balance analysis module and a particle swarm optimization algorithm are adopted to calculate the amount of power unbalance, and a new energy power consumption and energy storage adjustment prediction result is obtained. The charge and discharge curve of the energy storage system is predicted by using the LSTM model, and the adjustment flexibility is evaluated, so that the technical problems of reducing the active unbalance, reducing the abandoned electric quantity, predicting the charge and discharge requirements of the energy storage system and improving the operation efficiency of the energy storage system by combining the particle swarm optimization algorithm and the long short-term memory network are solved, the active unbalance is reduced, and the energy storage efficiency is improved. The electric power dispatching system is helped to reasonably dispatch energy storage resources, the operation efficiency of the energy storage system is improved, unnecessary electric energy waste is reduced, the flexibility and reliability of a power grid are improved, the electric power dispatching process is optimized, and the dispatching accuracy and economical efficiency are improved.

Description

Technical Field

[0001] The present invention belongs to the field of electric power technology, and in particular relates to a new energy power consumption and energy storage regulation prediction method based on machine learning. Background Art

[0002] The rapid development of renewable energy such as wind power and solar energy has provided important support for achieving the global "dual carbon" goals. However, as the proportion of these new energy sources in the power system continues to increase, their volatility and intermittent characteristics have brought huge challenges to the stability and economy of the power system. Traditional power system scheduling relies on stable base load power sources, such as thermal power units, but with the increase in the proportion of new energy sources such as wind power and solar energy, traditional power sources are gradually replaced, and the regulation capacity of the power generation side has dropped significantly. Therefore, it is necessary to introduce more flexible and efficient regulation resources to ensure the stable operation of the power system.

[0003] As an important flexible adjustment method, energy storage system can balance the supply and demand of the power grid, alleviate the volatility of renewable energy power generation, and improve the regulation capacity of the power grid. However, the reasonable configuration and efficient use of energy storage system are crucial to the stable operation of the power system. At present, there are still many technical difficulties in the reasonable prediction of energy storage capacity, scheduling optimization and coordinated operation of the power grid.

[0004] Existing studies often ignore the comprehensive consideration of multi-dimensional constraints and the impact of real-time scheduling on system stability when applying these technologies. Most existing power dispatch methods rely on static optimization models, lack consideration of the dynamic response and real-time dispatch capabilities of renewable energy generation, and are unable to meet the dispatch needs of power systems under the condition of high penetration of renewable energy. Summary of the invention

[0005] The purpose of the present invention is to provide a new energy power consumption and energy storage regulation prediction method based on machine learning, which solves the technical problem of reducing active power imbalance, reducing power abandonment, predicting the charging and discharging demand of the energy storage system, helping the power dispatching system to reasonably dispatch energy storage resources, and improving the operating efficiency of the energy storage system by combining the particle swarm optimization algorithm with the long short-term memory network LSTM.

[0006] To achieve the above object, the present invention adopts the following technical solution:

[0007] A new energy power consumption and energy storage regulation prediction method based on machine learning includes the following steps:

[0008] Step 1: Establish a data acquisition module cluster at the data acquisition layer, and obtain power grid and energy storage data from data sources in the power grid system through the Internet, including power grid topology and operating parameters, predicted output curves and actual output curves of new energy stations, energy storage system charging and discharging capacity parameters, day-ahead load forecast curves, and historical regulation data;

[0009] Step 2: The data processing and storage layer obtains the power grid and energy storage data of the data acquisition layer, pre-processes the power grid and energy storage data through the pre-processing module, obtains the original data, and builds the original data database;

[0010] Step 3: The analysis and decision-making layer retrieves the raw data from the raw data database and analyzes the raw data according to the following steps:

[0011] Step 3-1: Establish an active power balance analysis module, retrieve the original data, construct the economic dispatch objective function and constraints with the lowest cost, calculate and analyze the power imbalance, use the particle swarm optimization algorithm PSO to find the optimal solution, and output the global optimal solution and power imbalance analysis results;

[0012] Step 3-2: Establish an energy storage capacity adjustment prediction module, obtain the output results and original data of step 1, use the LSTM long short-term memory network model to predict the energy storage charging and discharging curve of the energy storage system in the future scheduling cycle, evaluate the adjustment flexibility and adjustment time of the energy storage, obtain the adjustment flexibility index, and output the energy storage charging and discharging curve and adjustment flexibility index;

[0013] Step 4: The user interaction and visualization layer obtains the output results of the analysis and decision-making layer, and builds a visualization module to display the global optimal solution, power imbalance analysis results, energy storage charging and discharging curves, and adjustment flexibility indicators.

[0014] Preferably, at the data acquisition layer, the data acquisition module cluster includes a power grid data acquisition module, a new energy station data acquisition module, an energy storage system data acquisition module, a load forecasting data acquisition module, and a historical dispatching data acquisition module;

[0015] The data sources include power grid control system, power grid dispatch center, new energy station prediction system, energy storage management system and load prediction system;

[0016] The power grid data acquisition module is responsible for obtaining the power grid topology and operating parameters from the power grid control system through the Internet;

[0017] The new energy station data acquisition module is responsible for obtaining the predicted output curve and actual output curve of the new energy station from the new energy station prediction system through the Internet;

[0018] The energy storage system data acquisition module is responsible for obtaining the energy storage system charging and discharging capacity parameters from the energy storage management system through the Internet, including the energy storage system's charging power limit, discharging power limit, and charging and discharging efficiency;

[0019] The load forecast data acquisition module is responsible for obtaining the day-ahead load forecast curve from the load forecast system through the Internet;

[0020] The historical dispatch data acquisition module is responsible for obtaining historical regulation data from the power grid dispatch center through the Internet, including historical load, generator power, and energy storage power regulation data.

[0021] Preferably, the power grid control system and the new energy station prediction system are both SCADA systems.

[0022] Preferably, when executing step 3-1, the specific steps are as follows:

[0023] Step 3-1-1: Construct the economic dispatch objective function and constraints with the minimum cost. The specific formula is as follows:

[0024] min{C G +ρE wcur +C battery -εE tie +γE Loadcur}.

[0025] Among them, C G is the cost of purchasing electricity from the power plant; E wcur is the amount of power wasted; C battery is the operating cost of the energy storage system; E tie is the amount of electricity delivered; E Loadcur The amount of lost load; ρ is the penalty coefficient for power abandonment, in RMB / kWh, ∈ is the profit coefficient of the power transmission, in RMB / kWh, and γ is the penalty coefficient for lost load, in RMB / kWh;

[0026]

[0027] Among them, C gen，i is the unit power generation cost of the i-th generator, in yuan / kWh; P gen，i (t) is the output power of the ith generator at time t, in MW; N gen is the total number of generators;

[0028]

[0029] in, is the predicted output of the i-th renewable energy station at time i, in MW, is the actual output of the i-th renewable energy station at time t, in MW; Nrenewable is the total number of new energy stations; Δt is the time step;

[0030]

[0031] Among them, C charge is the unit cost of energy storage charging, in yuan / kWh, C discharge is the unit cost of energy storage discharge, in yuan / kWh, is the total charging power of the energy storage system at time t, in MW, is the total discharge power of the energy storage system at time t, in MW;

[0032]

[0033] Among them, P tie (t) is the power delivered by the power grid, in MW; T is the number of time steps;

[0034]

[0035] Among them, P load (t) is the load demand at time t, in MW; t ranges from 1 to T; P battery，i (t) is the output power of the i-th energy storage device at time t;

[0036] Constraints:

[0037] Power balance constraints:

[0038]

[0039] P battery (t) is the total power output of the energy storage system at time t;

[0040] E wcur (t) is the amount of power abandoned at time t:

[0041]

[0042] E wcur The constraint of (t) is: E wcur (t)≥0;

[0043] Energy storage power limit:

[0044] -P battery，max ≤P battery (t)≤P battery，max ;

[0045] Among them, -P battery，max Indicates the maximum power limit of the energy storage system discharge, P battery，maxIndicates the maximum power limit of the energy storage system charging; P battery (t) is the power output of the energy storage system at time t;

[0046] Generator power limit:

[0047] P gen，i，mi n≤P gen，i (t)≤P gen，i，max ;

[0048] Among them, P gen，i，min represents the minimum power output limit of the ith generator set, P gen，i，max represents the maximum power output limit of the i-th generator set;

[0049] Step 3-1-2: Use the particle swarm optimization algorithm PSO to find the optimal solution, which includes the following steps:

[0050] Step 3-1-2-1: Initialize the number of particles in the particle swarm and randomly generate a position in the solution space for each particle. The position of each particle represents a scheduling solution. The position of the particle contains multiple variables, including P gen，i (t) and P battery，i (t);

[0051] Step 3-1-2-2: For each particle, calculate the value of the objective function, which is expressed as:

[0052] f(x)=C gen (x)+ρE cur (x)+C battery (x)-∈E tie (x)+γE loadcur (x);

[0053] Step 3-1-2-3: According to the PSO algorithm, each particle updates its position and speed based on its own historical best solution and the historical best solution of all particles:

[0054]

[0055] in, is the velocity of particle i at the tth iteration; is the position of particle i at the tth iteration; w is the inertia weight; c1 and c2 are acceleration constants, which control the speed at which the particle approaches the individual optimal solution pbest and the global optimal solution gbest; r1 and r2 are random numbers with a value range of [0, 1]; pbest i is the individual optimal solution of particle i; gbest is the global optimal solution of all particles;

[0056] Step 3-1-2-4: Constraint processing, specifically, if the solution of the particle violates the constraint condition, a penalty term is added to the solution to increase the objective function value and reduce the priority in the search process;

[0057] Step 3-1-2-5: Stop condition, specifically reaching the maximum number of iterations or convergence of the global optimal solution;

[0058] Step 3-1-3: Output results, including output of global optimal solution and power imbalance analysis results;

[0059] The global optimal solution is the optimal plan for grid dispatch, including the output plan of each motor and the charging and discharging plan of each energy storage system;

[0060] The power imbalance analysis results include: Loadcur >0, it indicates that the peak-shaving capacity is insufficient or there is an imbalance between supply and demand; otherwise, there is no imbalance.

[0061] Preferably, when executing step 3-2, the following steps are specifically included:

[0062] Step 3-2-1: The input data includes the output result obtained from step 3-1 and the original data;

[0063] Step 3-2-2: Construct an LSTM model based on time series prediction. The input layer includes the predicted output of new energy, load prediction, historical regulation data, charging and discharging capacity parameters of the energy storage system, the output plan of each generator, and the charging and discharging plan of the energy storage system; the hidden layer includes extracting time series features through multi-layer LSTM units; the output layer includes predicting the charging and discharging power of the energy storage system at future times;

[0064] Step 3-2-3: Use historical adjustment data to train the LSTM model and optimize model parameters;

[0065] Step 3-3-4: Output of prediction results, including energy storage charging and discharging curves and adjustment flexibility indicators;

[0066] The energy storage charge and discharge curve is used to predict the charge and discharge power curve of the energy storage system at future times;

[0067] The regulation flexibility index is used to evaluate the regulation capability of the energy storage system in different time periods, including the maximum response time of charging / discharging and flexibility index.

[0068] The method for predicting the adjustment of new energy power consumption and energy storage based on machine learning described in the present invention solves the technical problems of reducing active power imbalance, reducing abandoned power, predicting the charging and discharging demand of energy storage system, helping power dispatching system to reasonably dispatch energy storage resources, and improving the operation efficiency of energy storage system by combining particle swarm optimization algorithm with long short-term memory network LSTM, and optimizing the active power balance of power grid through particle swarm optimization (PSO) algorithm, which can quickly respond to new energy fluctuation and load change, reduce active power imbalance, reduce abandoned power, and improve the stability and economy of power grid, and accurately predict the charging and discharging demand of energy storage system by learning historical data through LSTM network, and help power dispatching system to reasonably dispatch energy storage resources, improve the operation efficiency of energy storage system, and reduce unnecessary waste of electric energy, and perform real-time dispatching through particle swarm optimization algorithm, which can timely respond to dynamic changes in power grid operation, quickly adjust power grid load and charging and discharging plan of energy storage system, and improve the flexibility and reliability of power grid, and comprehensively consider multi-dimensional constraints such as power grid topology, generator output, energy storage power limit, and power abandonment penalty fee, optimize the power dispatching process, and improve the accuracy and economy of dispatching. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is the main flow chart of the present invention;

[0070] Figure 2 is a flow chart of step 3-1 of the present invention;

[0071] Figure 3 is a flow chart of step 3-2 of the present invention;

[0072] Figure 4 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0073] Depend on Figure 1-Figure 4 A new energy power consumption and energy storage regulation prediction method based on machine learning is shown, comprising the following steps:

[0074] Step 1: Establish a data acquisition module cluster at the data acquisition layer, and obtain power grid and energy storage data from data sources in the power grid system through the Internet, including power grid topology and operating parameters, predicted output curves and actual output curves of new energy stations, energy storage system charging and discharging capacity parameters, day-ahead load forecast curves, and historical regulation data;

[0075] At the data acquisition layer, the data acquisition module cluster includes a power grid data acquisition module, a new energy station data acquisition module, an energy storage system data acquisition module, a load forecasting data acquisition module, and a historical dispatching data acquisition module;

[0076] The data sources include a power grid control system, a power grid dispatching center, a new energy site prediction system, an energy storage management system, and a load prediction system; the power grid control system, the power grid dispatching center, the new energy site prediction system, the energy storage management system, and the load prediction system are all existing technologies, and this embodiment obtains data in these systems through the Internet.

[0077] The power grid data acquisition module is responsible for obtaining the power grid topology and operating parameters from the power grid control system through the Internet;

[0078] The Grid Control System (GCS) is one of the core components of the power system, responsible for real-time monitoring, dispatching and fault handling of the power grid. It ensures the stable operation of the power grid by controlling the power grid equipment (such as substations, switches, generators, etc.).

[0079] In this embodiment, the power grid control system is usually based on the SCADA (Supervisory Control and Data Acquisition) system, which is a widely used remote monitoring and control technology that can obtain the operating status of the power grid (such as current, voltage, frequency, power and other parameters) in real time and perform necessary control operations.

[0080] The grid topology and operating parameters include the configuration and operating parameters of grid nodes, lines, power sources, loads, etc. The input data involved in the calculation in step 3 include:

[0081] P gen，i (t) the output power of each generator at time t;

[0082] P gen，i，min Minimum power output limit for each generator set;

[0083] P gen，i，max Maximum power output limit of each generator set;

[0084] P hattery (t) total output power of the energy storage system at time t;

[0085] P load (t) load demand at time t;

[0086] C gen，i The unit electricity cost of each generator;

[0087] P tie (t) Power exported from the power grid.

[0088] The new energy station data acquisition module is responsible for obtaining the predicted output curve and actual output curve of the new energy station from the new energy station prediction system through the Internet;

[0089] In this embodiment, the predicted output curve and the actual output curve are the power output data of new energy sites such as wind power and photovoltaic power, including predicted data and actual operation data. The input data involved in the calculation in step 3 include:

[0090] The predicted output of each renewable energy station at time i;

[0091] The actual output of each renewable energy station at time t;

[0092] E wcure The amount of power curtailment in the calculation of curtailment cost;

[0093] The new energy station prediction system is mainly used to predict the power generation output of new energy (such as wind power and photovoltaic) stations. The power generation of new energy is affected by factors such as weather and environment, and has volatility and uncertainty. Therefore, the prediction system can provide the power grid dispatching center with a more accurate prediction of new energy power generation.

[0094] The new energy prediction system can be modeled based on meteorological data, historical output data and machine learning algorithms (such as neural networks, LSTM, etc.) to predict the output of new energy sites such as wind power and photovoltaics.

[0095] The energy storage system data acquisition module is responsible for obtaining the energy storage system charging and discharging capacity parameters from the energy storage management system through the Internet, including the energy storage system's charging power limit, discharging power limit, and charging and discharging efficiency. The input data involved in the calculation in step 3 include:

[0096] The total charging power of the energy storage system at time t;

[0097] The total discharge power of the energy storage system at time t;

[0098] C charge Unit cost of energy storage charging;

[0099] C discharge Unit cost of energy storage discharge;

[0100] -P battery，max Maximum power limit for energy storage system discharge;

[0101] P battery，max The maximum power limit for charging the energy storage system.

[0102] Energy Storage Management System (ESMS) is a technical system used to manage the operation and dispatch of large-scale energy storage facilities. It can maximize the efficiency and economic value of energy storage systems by intelligently controlling the charging and discharging process of energy storage systems.

[0103] Modern energy storage management systems integrate advanced monitoring, control and optimization scheduling functions. Energy storage systems include battery energy storage, flywheel energy storage and other forms. ESMS can monitor the charge and discharge status of batteries, manage the health status of battery packs, and dispatch according to factors such as grid demand and market prices.

[0104] The load forecast data acquisition module is responsible for obtaining the day-ahead load forecast curve from the load forecast system through the Internet;

[0105] The day-ahead load forecast curve is the forecast data of future load demand. The input data involved in the calculation in step 3 include:

[0106] E Loadcur Loss of load.

[0107] In this embodiment, the power grid dispatch center is equipped with an energy management system (EMS), which integrates load forecasting, power grid optimization dispatching, real-time monitoring and other functions, and is widely used in the operation and dispatching of power grids. EMS is not only responsible for dispatching, but also can analyze the load fluctuation of the power grid in real time and adjust the output of the generator set.

[0108] The grid dispatching center is the core unit in the power system responsible for coordinating various types of power resources and dispatching power generation, energy storage and load to meet the load demand of the power grid. It performs real-time load forecasting, power generation dispatching and optimized dispatching through the operation dispatching system.

[0109] At present, the load forecasting system can combine historical load data, weather factors, economic activities, etc. to build models. Commonly used algorithms include time series analysis (such as ARIMA), machine learning methods (such as XGBoost, LSTM), etc. Its purpose is to predict the load demand of the power grid in advance, so as to avoid the problem of insufficient or excessive power.

[0110] The load forecasting system is used to predict the changes in power grid load in the future and help the power grid dispatching center to reasonably arrange power generation and dispatching plans.

[0111] The historical dispatch data acquisition module is responsible for obtaining historical regulation data from the power grid dispatch center through the Internet, including historical load, generator power, and energy storage power regulation data. The input data involved in the calculation in step 3 include:

[0112] Historical data:

[0113] P gen，i (t) the output power of each generator at time t;

[0114] The total charging power of the energy storage system at time t;

[0115] The total discharge power of the energy storage system at time t;

[0116] P battery，i (t) Output power of each energy storage device at time t.

[0117] The power grid control system and the new energy station prediction system are both SCADA systems.

[0118] The present invention adopts a multi-layer architecture, including data collection, data processing, analysis and decision-making, and user interaction and visualization layers. This architecture supports efficient data flow and module collaboration.

[0119] Step 2: The data processing and storage layer obtains the power grid and energy storage data of the data acquisition layer, pre-processes the power grid and energy storage data through the pre-processing module, obtains the original data, and builds the original data database;

[0120] Preprocessing includes data cleaning, data normalization, data interpolation, etc.

[0121] Step 3: The analysis and decision-making layer retrieves the raw data from the raw data database and analyzes the raw data according to the following steps:

[0122] Step 3-1: Establish an active power balance analysis module, retrieve the original data, construct the economic dispatch objective function and constraints with the lowest cost, calculate and analyze the power imbalance, use the particle swarm optimization algorithm PSO to find the optimal solution, and output the global optimal solution and power imbalance analysis results;

[0123] When executing step 3-1, the specific steps are as follows:

[0124] Step 3-1-1: Construct the economic dispatch objective function and constraints with the minimum cost. The specific formula is as follows:

[0125] min{C G +ρE wcur +C battery -εE tie +γE Loadcur};

[0126] Among them, C G is the cost of purchasing electricity from the power plant; E wcur is the amount of power wasted; Cbattery is the operating cost of the energy storage system; E tie is the amount of electricity delivered; E Loadcur The amount of lost load; ρ is the penalty coefficient for power abandonment, in RMB / kWh, ∈ is the profit coefficient of the power transmission, in RMB / kWh, and γ is the penalty coefficient for lost load, in RMB / kWh;

[0127] In this embodiment, ρ represents the economic cost or penalty caused by power abandonment. The output of new energy (such as wind power and photovoltaic power) may not be fully absorbed (i.e., "abandoned") in some cases. In order to reflect the economic loss of power abandonment, this embodiment sets a power abandonment penalty coefficient ρ to calculate the additional cost caused by power abandonment. In the objective function, ρE wcur Represents the penalty cost caused by power curtailment.

[0128] ∈ is the revenue coefficient of the electricity delivered, ∈E tie It represents the benefits of exporting electricity to external power grids. If the power grid can export more electricity to other power grids or regions, then E tie will increase, thus bringing more benefits. Since it is in the form of a negative sign, it means that the increase in the amount of electricity sent out will reduce the total cost.

[0129] γ is the load loss penalty coefficient, γE loadcur represents the penalty cost of load loss. If the power supply of the power grid cannot meet the predicted load demand, E Loadur will increase, resulting in an increase in system cost. Load loss usually means unstable grid frequency or power outage, which needs to be avoided as much as possible in this embodiment.

[0130] In this embodiment, the objective function aims to optimize the economic dispatch of the power grid to minimize the operating cost of the power grid, while also balancing factors such as power abandonment, external power transmission, and load loss. By adjusting these penalty coefficients, the optimization algorithm can find the optimal power supply plan under various dispatch conditions to ensure the economy and stability of the power grid.

[0131] The power abandonment penalty coefficient ρ encourages reducing power abandonment and consuming new energy power as much as possible; the revenue coefficient ∈ of the transmitted power encourages the transmission of power and increases the economic benefits of the power grid; the load loss penalty coefficient γ punishes the situation where the power grid cannot meet the load demand and ensures the reliability of the power supply of the power grid.

[0132]

[0133] Among them, C gen，i is the unit power generation cost of the i-th generator, in yuan / kWh; P gen，i (t) is the output power of the ith generator at time t, in MW; N gen is the total number of generators;

[0134]

[0135] in, is the predicted output of the i-th renewable energy station at time t, in MW, is the actual output of the i-th renewable energy station at time t, in MW; N renewable is the total number of new energy stations; Δt is the time step;

[0136]

[0137] Among them, C charge is the unit cost of energy storage charging, in yuan / kWh, C discharge is the unit cost of energy storage discharge, in yuan / kWh, is the total charging power of the energy storage system at time t, in MW, is the total discharge power of the energy storage system at time t, in MW;

[0138]

[0139] Among them, P tie (t) is the power delivered by the power grid, in MW; T is the number of time steps;

[0140]

[0141] Among them, P load (t) is the load demand at time t, in MW; t ranges from 1 to T; P battery，i (t) is the output power of the i-th energy storage device at time t;

[0142] Constraints:

[0143] Power balance constraints:

[0144]

[0145] P battery (t) is the total power output of the energy storage system at time t;

[0146] E wcur (t) is the amount of power abandoned at time t:

[0147]

[0148] E wcur The constraint of (t) is: E wcur (t)≥0;

[0149] Energy storage power limit:

[0150] -Pbattery，max ≤P battery (i)≤P battery，max ;

[0151] Among them, -P battery，max Indicates the maximum power limit of the energy storage system discharge, P battert，max Indicates the maximum power limit of the energy storage system charging; P battery (t) is the power output of the energy storage system at time t;

[0152] Generator power limit:

[0153] P gen，i，min ≤P gen，i (t)≤P gen，i，max ;

[0154] Among them, P gen，imin represents the minimum power output limit of the ith generator set, P gen，i，max represents the maximum power output limit of the i-th generator set;

[0155] Step 3-1-2: Use the particle swarm optimization algorithm PSO to find the optimal solution, which includes the following steps:

[0156] Step 3-1-2-1: Initialize the number of particles in the particle swarm and randomly generate a position in the solution space for each particle. The position of each particle represents a scheduling solution. The position of the particle contains multiple variables, including P gen，i (t) and P battery，i (t);

[0157] Step 3-1-2-2: For each particle, calculate the value of the objective function, which is expressed as:

[0158] f(x)=C gen (x)+ρE cur (x)+C battery (x)-∈E ti e(x)+γE loadcur (x);

[0159] Step 3-1-2-3: According to the PSO algorithm, each particle updates its position and speed based on its own historical best solution and the historical best solution of all particles:

[0160]

[0161] in, is the velocity of particle i at the tth iteration; is the position of particle i at the tth iteration; w is the inertia weight; c1 and c2 are acceleration constants, which control the speed at which the particle approaches the individual optimal solution pbest and the global optimal solution gbest; r1 and r2 are random numbers with a value range of [0, 1]; pbest i is the individual optimal solution of particle i; gbest is the global optimal solution of all particles;

[0162] Step 3-1-2-4: Constraint processing, specifically, if the solution of the particle violates the constraint condition, a penalty term is added to the solution to increase the objective function value and reduce the priority in the search process;

[0163] Step 3-1-2-5: Stop condition, specifically reaching the maximum number of iterations or convergence of the global optimal solution;

[0164] Step 3-1-3: Output results, including output of global optimal solution and power imbalance analysis results;

[0165] The global optimal solution is the optimal plan for grid dispatch, including the output plan of each generator and the charging and discharging plan of each energy storage system;

[0166] In this embodiment, the output plan of the motor includes the power output plan of each generator in each time period;

[0167] The charging and discharging plan of the energy storage system includes the charging and discharging plans of each energy storage system in each time period.

[0168] The power imbalance analysis result is E Loadcur , including if E Loadcur >0, it indicates that the peak-shaving capacity is insufficient or there is an imbalance between supply and demand; otherwise, there is no imbalance.

[0169] Step 3-2: Establish an energy storage capacity adjustment prediction module, obtain the output results and original data of step 1, use the LSTM long short-term memory network model to predict the energy storage charging and discharging curve of the energy storage system in the future scheduling cycle, evaluate the adjustment flexibility and adjustment time of the energy storage, obtain the adjustment flexibility index, and output the energy storage charging and discharging curve and adjustment flexibility index;

[0170] When executing step 3-2, the specific steps include:

[0171] Step 3-2-1: The input data includes the output result obtained from step 3-1 and the original data;

[0172] Step 3-2-2: Construct an LSTM model based on time series prediction. The input layer includes the predicted output of new energy, load prediction, historical regulation data, charging and discharging capacity parameters of the energy storage system, the output plan of each generator, and the charging and discharging plan of the energy storage system; the hidden layer includes extracting time series features through multi-layer LSTM units; the output layer includes predicting the charging and discharging power of the energy storage system at future times;

[0173] In this embodiment, the time series features of the input layer input of the LSTM model include:

[0174] Output plan of each generator;

[0175] The charging and discharging plan of the energy storage system;

[0176] Power imbalance analysis results E Loadcur ;

[0177] The predicted output of the i-th renewable energy station at time t;

[0178] P load (t) load demand at time t;

[0179] P in historical data gen，i (t) The output power of generator i at time t and P battry，i (t) The output power of the i-th energy storage device at time t;

[0180] -P battery，max Maximum power limit for energy storage system discharge;

[0181] P battery，max The maximum power limit for charging the energy storage system.

[0182] The hidden layer extracts time series features through multi-layer LSTM units to capture the temporal dependencies between different moments and different inputs;

[0183] The output layer outputs the charging and discharging power of the energy storage system in the future scheduling cycle, including the energy storage charging and discharging curve It represents the predicted charging and discharging power of each energy storage system i at time t; the adjustment flexibility index, including the maximum response time and flexibility index, is used to evaluate the adjustment capability of the energy storage system, that is, the charging / discharging capability and duration of the energy storage system when facing different loads or fluctuations in the output of new energy.

[0184] Step 3-2-3: Use historical adjustment data to train the LSTM model and optimize model parameters;

[0185] In this embodiment, the LSTM model is trained using historical adjustment data to optimize the parameters of the model (e.g., LSTM weights and biases). The goal of the training is to enable the model to predict how the energy storage system responds and the charging and discharging strategy of the energy storage under different grid loads and new energy output conditions.

[0186] The objective function during training is:

[0187]

[0188] in, is the actual charge and discharge power in the historical data, is the charge and discharge power predicted by the LSTM model.

[0189] Step 3-3-4: Output of prediction results, including energy storage charging and discharging curves and adjustment flexibility indicators;

[0190] The energy storage charge and discharge curve is used to predict the charge and discharge power curve of the energy storage system at future times;

[0191] The regulation flexibility index is used to evaluate the regulation capability of the energy storage system in different time periods, including the maximum response time of charging / discharging and flexibility index.

[0192] Step 4: The user interaction and visualization layer obtains the output results of the analysis and decision-making layer, and builds a visualization module to display the supply and demand balance status of the power grid, the energy storage regulation capability and energy storage charging and discharging curves, the generator output and scheduling plan, and the power imbalance analysis.

[0193] In this embodiment, the data content displayed by the user interaction and visualization layer is displayed in the form of charts, including:

[0194] The supply and demand balance of the power grid: including the display of the current power grid load P load (t) The gap between the power supply capacity of the grid (i.e. the power provided by all generators and energy storage systems);

[0195] Display the amount of abandoned electricity E wcur .

[0196] Energy storage regulation capability and energy storage charge and discharge curve: including display of energy storage charge and discharge curve within the forecast period

[0197] Display regulation flexibility indicators, such as maximum response time of charging / discharging, flexibility index, etc.;

[0198] Display-P battery，max The maximum power limit of the energy storage system discharge and P battery，max Maximum power limit for charging the energy storage system;

[0199] Display P battery (t) the output power of the energy storage system at that moment;

[0200] Display the charging and discharging plan of the overall energy storage system.

[0201] Generator output and dispatch plan: including the display of the overall generator output plan.

[0202] Power imbalance analysis: including display of power imbalance of the power grid (load loss E Loadcur ).

[0203] In this embodiment, the data collection layer, data processing and storage layer, analysis and decision-making layer, and user interaction and visualization layer are all deployed in the cloud server and communicate with each other through the Internet.

[0204] The method for predicting the adjustment of new energy power consumption and energy storage based on machine learning described in the present invention solves the technical problems of reducing active power imbalance, reducing abandoned power, predicting the charging and discharging demand of energy storage system, helping power dispatching system to reasonably dispatch energy storage resources, and improving the operation efficiency of energy storage system by combining particle swarm optimization algorithm with long short-term memory network LSTM, and optimizing the active power balance of power grid through particle swarm optimization (PSO) algorithm, which can quickly respond to new energy fluctuation and load change, reduce active power imbalance, reduce abandoned power, and improve the stability and economy of power grid, and accurately predict the charging and discharging demand of energy storage system by learning historical data through LSTM network, and help power dispatching system to reasonably dispatch energy storage resources, improve the operation efficiency of energy storage system, and reduce unnecessary waste of electric energy, and perform real-time dispatching through particle swarm optimization algorithm, which can timely respond to dynamic changes in power grid operation, quickly adjust power grid load and charging and discharging plan of energy storage system, and improve the flexibility and reliability of power grid, and comprehensively consider multi-dimensional constraints such as power grid topology, generator output, energy storage power limit, and power abandonment penalty fee, optimize the power dispatching process, and improve the accuracy and economy of dispatching.

Claims

1. A new energy power consumption and energy storage regulation prediction method based on machine learning, characterized by: The steps include: Step 1: Establish a data acquisition module cluster at the data acquisition layer, and obtain power grid and energy storage data from data sources in the power grid system through the Internet, including power grid topology and operating parameters, predicted output curves and actual output curves of new energy stations, energy storage system charging and discharging capacity parameters, day-ahead load forecast curves, and historical regulation data; Step 2: The data processing and storage layer obtains the power grid and energy storage data of the data acquisition layer, pre-processes the power grid and energy storage data through the pre-processing module, obtains the original data, and builds the original data database; Step 3: The analysis and decision-making layer retrieves the raw data from the raw data database and analyzes the raw data according to the following steps: Step 3-1: Establish an active power balance analysis module, retrieve the original data, construct the economic dispatch objective function and constraints with the lowest cost, calculate and analyze the power imbalance, use the particle swarm optimization algorithm PSO to find the optimal solution, and output the global optimal solution and power imbalance analysis results; Step 3-2: Establish an energy storage capacity adjustment prediction module, obtain the output results and original data of step 1, use the LSTM long short-term memory network model to predict the energy storage charging and discharging curve of the energy storage system in the future scheduling cycle, evaluate the adjustment flexibility and adjustment time of the energy storage, obtain the adjustment flexibility index, and output the energy storage charging and discharging curve and adjustment flexibility index; Step 4: The user interaction and visualization layer obtains the output results of the analysis and decision-making layer, and builds a visualization module to display the global optimal solution, power imbalance analysis results, energy storage charging and discharging curves, and adjustment flexibility indicators.

2. The method for predicting the consumption and storage regulation of new energy power based on machine learning according to claim 1, characterized in that: At the data acquisition layer, the data acquisition module cluster includes a power grid data acquisition module, a new energy station data acquisition module, an energy storage system data acquisition module, a load forecasting data acquisition module, and a historical dispatching data acquisition module; The data sources include power grid control system, power grid dispatch center, new energy station prediction system, energy storage management system and load prediction system; The power grid data acquisition module is responsible for obtaining the power grid topology and operating parameters from the power grid control system through the Internet; The new energy station data acquisition module is responsible for obtaining the predicted output curve and actual output curve of the new energy station from the new energy station prediction system through the Internet; The energy storage system data acquisition module is responsible for obtaining the energy storage system charging and discharging capacity parameters from the energy storage management system through the Internet, including the energy storage system's charging power limit, discharging power limit, and charging and discharging efficiency; The load forecast data acquisition module is responsible for obtaining the day-ahead load forecast curve from the load forecast system through the Internet; The historical dispatch data acquisition module is responsible for obtaining historical regulation data from the power grid dispatch center through the Internet, including historical load, generator power, and energy storage power regulation data.

3. The method for predicting the consumption and storage regulation of new energy power based on machine learning according to claim 1, characterized in that: The power grid control system and the new energy station prediction system are both SCADA systems.

4. The method for predicting the consumption and storage regulation of new energy power based on machine learning according to claim 1, characterized in that: When executing step 3-1, the specific steps are as follows: Step 3-1-1: Construct the economic dispatch objective function and constraints with the minimum cost. The specific formula is as follows: min{C G +ρE wcur +C battery -εE tie +γE Loader }; Among them, C G E is the cost of purchasing electricity from the power plant; wcur is the amount of power wasted; C battery is the operating cost of the energy storage system; E tie is the amount of electricity delivered; E Loadcur The amount of lost load; ρ is the penalty coefficient for power abandonment, in RMB / kWh, ∈ is the profit coefficient of the power transmission, in RMB / kWh, and γ is the penalty coefficient for lost load, in RMB / kWh; Among them, C gen，i is the unit power generation cost of the i-th generator, in yuan / kWh; P gen，i (t) is the output power of the ith generator at time t, in MW; N gen is the total number of generators; in, is the predicted output of the i-th renewable energy station at time t, in MW, is the actual output of the i-th renewable energy station at time t, in MW; N renewable is the total number of new energy stations; Δt is the time step; Among them, C charge is the unit cost of energy storage charging, in yuan / kWh, C discharge is the unit cost of energy storage discharge, in yuan / kWh, is the total charging power of the energy storage system at time t, in MW, is the total discharge power of the energy storage system at time t, in MW; Among them, P tie (t) is the power delivered by the power grid, in MW; T is the number of time steps; Among them, P load (t) is the load demand at time t, in MW; t ranges from 1 to T; P battery，i (t) is the output power of the i-th energy storage device at time t; Constraints: Power balance constraints: P battery (t) is the total power output of the energy storage system at time t; E wcur (t) is the amount of power abandoned at time t: E wcur The constraint of (t) is: E wcur (t)≥0; Energy storage power limit: -P battery，max ≤P battery (t)≤P battery，max ; Among them, -P battery，max Indicates the maximum power limit of the energy storage system discharge, P battery，max Indicates the maximum power limit of the energy storage system charging; P battery (t) is the power output of the energy storage system at time t; Generator power limit: P gen，i，min ≤P gen，i (t)≤P gen,i，max ; Among them, P gen，i，min represents the minimum power output limit of the ith generator set, P gen，i，max represents the maximum power output limit of the i-th generator set; Step 3-1-2: Use the particle swarm optimization algorithm PSO to find the optimal solution, which includes the following steps: Step 3-1-2-1: Initialize the number of particles in the particle swarm and randomly generate a position in the solution space for each particle. The position of each particle represents a scheduling solution. The position of the particle contains multiple variables, including P gen，i (t) and P battery，i (t); Step 3-1-2-2: For each particle, calculate the value of the objective function, which is expressed as: f(x)=C gen (x)+ρE cur (x)+C battery (x)-∈E tie (x)+γE loadcur (x); Step 3-1-2-3: According to the PSO algorithm, each particle updates its position and speed based on its own historical best solution and the historical best solution of all particles: in, is the velocity of particle i at the tth iteration; is the position of particle i at the tth iteration; w is the inertia weight; c1 and c2 are acceleration constants, which control the speed at which the particle approaches the individual optimal solution pbest and the global optimal solution gbest; r1 and r2 are random numbers with a value range of [0,1]; pbesti is the individual optimal solution of particle i; gbest is the global optimal solution of all particles; Step 3-1-2-4: Constraint processing, specifically, if the solution of the particle violates the constraint condition, a penalty term is added to the solution to increase the objective function value and reduce the priority in the search process; Step 3-1-2-5: Stop condition, specifically reaching the maximum number of iterations or convergence of the global optimal solution; Step 3-1-3: Output results, including output of global optimal solution and power imbalance analysis results; The global optimal solution is the optimal plan for grid dispatch, including the output plan of each motor and the charging and discharging plan of each energy storage system; The power imbalance analysis results include: Loadcur >0, it indicates insufficient peak-shaving capacity or imbalance between supply and demand; otherwise, there is no imbalance.

5. The method for predicting the consumption and storage regulation of new energy power based on machine learning according to claim 4, characterized in that: When executing step 3-2, the specific steps include: Step 3-2-1: The input data includes the output result obtained from step 3-1 and the original data; Step 3-2-2: Construct an LSTM model based on time series prediction. The input layer includes the predicted output of new energy, load prediction, historical regulation data, charging and discharging capacity parameters of the energy storage system, the output plan of each generator, and the charging and discharging plan of the energy storage system; the hidden layer includes extracting time series features through multi-layer LSTM units; the output layer includes predicting the charging and discharging power of the energy storage system at future times; Step 3-2-3: Use historical adjustment data to train the LSTM model and optimize model parameters; Step 3-3-4: Output of prediction results, including energy storage charging and discharging curves and adjustment flexibility indicators; The energy storage charge and discharge curve is used to predict the charge and discharge power curve of the energy storage system at future times; The regulation flexibility index is used to evaluate the regulation capability of the energy storage system in different time periods, including the maximum response time of charging or discharging and flexibility index.

Images