Energy storage scheduling method and system based on service area load and renewable energy power generation prediction

Abstract

The invention discloses an energy storage scheduling method and system based on a service area load and renewable energy power generation prediction, and the method carries out the prediction of the load demand of a service area based on a long short-term memory (LSTM) network, and combines the renewable energy power generation prediction with a power grid electricity price signal to dynamically adjust the charging and discharging strategy of an energy storage unit. By predicting the fluctuation of the load in the service area and the power generation condition of the renewable energy source, the optimal energy storage scheduling scheme can be automatically generated in different time periods, and the method is particularly suitable for scenes with large traffic flow fluctuation and complex power demand in the service area. According to the method, the dependence of the service area on the external power grid is effectively reduced, and the stability and economic benefits of system operation are improved.

Description

Technical Field

[0001] The present invention relates to power systems and energy management technologies, and in particular to an energy storage scheduling method and system based on service area load and renewable energy power generation prediction. Background Art

[0002] With the development of renewable energy, the application of energy storage technology in power systems is becoming increasingly important. Traditional energy storage dispatching systems are mostly used in industrial parks, urban power grids and other scenarios, mainly to solve the time mismatch between power generation and load demand. In the existing technology, load forecasting and energy storage dispatching methods have been widely studied in these application scenarios, but there are still significant differences and challenges for the scenarios of highway service areas.

[0003] The load demand of highway service areas has the following particularities: On the one hand, the electricity demand of service areas is closely related to traffic flow and has obvious periodic fluctuation characteristics. For example, during holidays, weekends or peak hours, traffic flow increases significantly, resulting in a significant increase in load demand; while at night or during off-peak hours, load demand decreases significantly. This uncertainty in electricity load makes it difficult for traditional systems based on fixed load forecasting and scheduling to cope with it. In addition, the load volatility of service areas is also affected by external factors such as weather and regional traffic control, making load forecasting and scheduling more complicated.

[0004] On the other hand, the energy supply in the service area is increasingly dependent on renewable energy, especially photovoltaic power generation. However, the randomness and volatility of photovoltaic power generation make it difficult to predict its power generation stably, which further increases the difficulty of scheduling the energy storage system. Therefore, how to perform real-time dynamic scheduling based on actual load demand and power generation to avoid energy waste or over-reliance on the power grid has become a major problem in existing technologies.

[0005] Existing load forecasting and energy storage scheduling systems usually only consider historical electricity consumption data and lack optimization for targeted scenarios, especially in service areas where load changes drastically, and cannot meet the needs of refined energy management. At the same time, many scheduling methods only consider load forecasting, ignoring the instability of photovoltaic power generation and the impact of electricity price fluctuations on system operating costs. Therefore, it is difficult for existing technologies to achieve intelligent and economical energy storage scheduling solutions in service area scenarios. Summary of the invention

[0006] Purpose of the invention: The purpose of the present invention is to provide a method and system for energy storage scheduling based on service area load and renewable energy power generation forecasts, which can predict the supply and demand balance in future time periods according to the load demand of the service area and the real-time situation of photovoltaic and wind power generation, and dynamically adjust the charging and discharging strategy of the energy storage unit to achieve efficient utilization of energy and balanced operation of the system.

[0007] Technical solution: The energy storage scheduling method based on service area load and renewable energy generation prediction of the present invention includes:

[0008] Data collection: Real-time collection of power generation, historical load data, and external characteristic data of power generation entities in the service area, where the external characteristic data includes weather data, traffic flow data, and time characteristics;

[0009] Load forecasting: Based on the historical load data and external characteristic data collected in real time, the long short-term memory network model is used to predict the load demand in the service area in the future period;

[0010] Power generation forecast: Based on weather data, use the multivariate linear regression model to predict the photovoltaic power generation in the future period, or use the wind speed time series model to predict the future wind power generation; the photovoltaic power generation or wind power generation is determined by the power generation entity;

[0011] Energy storage dispatch strategy generation: Based on the results of load forecast and power generation forecast, calculate the supply and demand balance value in the future period, judge the difference between power generation and load demand based on the supply and demand balance value, and formulate the charging and discharging strategy of the energy storage unit based on the difference between power generation and load demand; dynamically adjust the power state of the energy storage unit according to the current power state of the energy storage unit and the charging and discharging strategy to ensure that the power is within the allowable range; when the power of the energy storage unit reaches the upper or lower limit, decide whether to interact with the power grid, including purchasing electricity from the power grid or feeding electricity to the power grid, according to the electricity price signal of the external power grid, to form the final energy storage dispatch strategy to optimize energy utilization and economic benefits.

[0012] Furthermore, based on the historical load data and external characteristic data collected in real time, the long short-term memory network model is used to predict the load demand in the service area in the future period, including:

[0013] Data preprocessing: preprocess historical load data and external characteristic data to ensure data quality and consistency;

[0014] Model training: Use the pre-processed historical load data and external feature data to train the pre-built long short-term memory network model. During the training process, the model learns the dynamic relationship in the time series through the historical load data and external feature data to optimize the prediction accuracy;

[0015] Load forecasting: After model training is completed, the latest historical load data and external feature data are input into the long short-term memory network model. The long short-term memory network model generates the load forecast value of the service area in the future period through recursive calculation;

[0016] Result output and evaluation: Output the load forecast value of the service area in the future period predicted by the long short-term memory network model, and compare it with the actual value to evaluate the prediction accuracy.

[0017] Furthermore, the pre-processed historical load data and external feature data are used to train the pre-built long short-term memory network model. During the training process, the model learns the dynamic relationship in the time series through the historical load data and external feature data to optimize the prediction accuracy, including:

[0018] Data input: During the model training phase, data is input to the pre-built LSTM network model in the form of a time series window, which includes historical load data and external feature data;

[0019] Forward propagation: Forward propagation is calculated progressively by multiple time series windows, and the output h of each time series window is t Depends on the current input x t and the hidden state h at the previous moment t-1 ; The memory unit in the LSTM network model is used to store information. The memory unit is transmitted along the entire sequence, and the information is selectively saved or discarded through the gating mechanism, so that the LSTM network model can store useful load data features for a long time while avoiding interference from irrelevant information;

[0020] Loss function calculation: The goal of model training is to minimize the error between the predicted value and the true value. The error between the predicted value and the true value is calculated using the loss function.

[0021] Back propagation and weight update: After forward propagation and calculation of the loss function, the error gradient is fed back to the parameters of each layer through the back propagation algorithm, and the weight matrix and bias term are adjusted to minimize the loss function; among them, the back propagation algorithm uses the gradient descent method for optimization.

[0022] Furthermore, the forward propagation includes:

[0023] Forget gate calculation: Forget gate f t Control the information to be forgotten at the current moment to ensure that the network does not excessively remember irrelevant information; filter out historical load data that is irrelevant to future predictions through the forget gate:

[0024] f t =σ(W f ·[h t-1 , x t ]+b f )

[0025] Where σ represents the activation function used to control the input, forget and output gates; W f represents the weight matrix of the forget gate; h t-1represents the hidden state of the previous time step; x t Indicates the current input; b f Represents the bias term of the forget gate;

[0026] Input gate calculation: input gate i t It is used to selectively update the current input information to the memory unit. The input gate i t Allows capturing new load variation patterns by selectively updating the memory state:

[0027] i t =σ(W i ·[h t-1 , x t ]+b i )

[0028] Where W i represents the weight matrix of the input gate; b i represents the bias term of the input gate;

[0029] Candidate memory cell C′ t renew:

[0030] C′ t =tan h(W C ·[h t-1 , x t ]+b C )

[0031] In the formula, C′ t represents the candidate new cell state; W C represents the weight matrix of cell update; b C Represents the bias term for cell updating; tanh represents the hyperbolic tangent function, which is usually used to limit the range of output values ​​to (-1, 1);

[0032] Memory cell update: This step generates new candidate cell states to update the contents of the main memory cell. The formula is as follows:

[0033] C t =f t ·C t-1 +i t ·C′ t

[0034] In the formula, C t represents the cell state at the current time t; C t-1 Indicates the cell state at the previous moment;

[0035] Output gate calculation: output gate o t Control the cell state C at the current time t t The information that needs to be output is as follows:

[0036] o t =σ(W o ·[h t-1 , x t ]+b o )

[0037] In the formula, o t Represents the output gate value at the current moment, controlling the influence of the current cell state on the output at the next moment; W o represents the weight matrix of the output gate; b o Represents the bias term of the output gate;

[0038] The final hidden state h t Calculated based on the output gate and memory unit:

[0039] h t =o t ·tan h(C t )

[0040] In the formula, h t Represents the hidden state at the current moment, that is, the output of the LSTM unit at the current moment;

[0041] Through multiple layers of calculation, the hidden state of each time step is generated, and all hidden states are used to predict future loads.

[0042] Furthermore, the goal of model training is to minimize the error between the predicted value and the true value. The error between the predicted value and the true value is calculated using the loss function, including:

[0043] The loss function uses mean square error, which is calculated as follows:

[0044]

[0045] In the formula, L(θ) represents the loss function; θ represents the model parameters; N represents the number of training samples; t represents the current time; τ represents the predicted time step; P′ load (t+τ) represents the load value predicted by simulation; P load (t+τ) represents the actual load value.

[0046] Furthermore, the back propagation algorithm is optimized using the gradient descent method, and the update formula is:

[0047]

[0048] In the formula, W represents the weight matrix; η represents the learning rate, which controls the step size of each update; Represents the gradient of the loss function with respect to the weights.

[0049] Furthermore, after the model training is completed, the latest historical load data and external feature data are input into the long short-term memory network model. The long short-term memory network model generates the load forecast value of the service area in the future period through recursive calculation, including:

[0050] P′ load (t+τ)=f LSTM [P load (tn), W(tn), T trafff c(tn)]

[0051] In the formula, t represents the current time; τ represents the predicted time step; P′ load (t+τ) represents the load value predicted by simulation; P load (tn) represents the electricity load demand at the past n time points; W(tn) represents the climate conditions at the past n time points; T trafffc (tn) represents the traffic flow of the service area at the past n time points. Further, based on weather data, a multivariate linear regression model is used to predict the photovoltaic power generation in the future period, or a wind speed time series model is used to predict the future wind power generation; wherein, the predicted photovoltaic power generation or wind power generation is determined according to the power generation subject, including:

[0052] If the service area uses a photovoltaic power generation system, the output of photovoltaic power generation is closely related to the light intensity and temperature. Based on the historical power generation data P PV (tn) and current weather data W(t), and predict future power generation through a multivariate linear regression model:

[0053] P′ PV (t+τ)=α 0 +α 1 ·S(t)+α 2 ·T(t)

[0054] In the formula, P′ PV (t+τ) represents the predicted photovoltaic power generation; α 0 , α 1 , α 2 represents the regression coefficient obtained through training data; S(t) represents solar irradiance; T(t) represents temperature;

[0055] If the service area uses wind energy as clean energy, the prediction of wind power generation depends on wind speed. It is assumed that the historical wind speed data is v wind (tn), predict the future wind speed through the wind speed time series model, and combine the power curve of the wind turbine to predict the future wind power generation:

[0056] P′ wind (t+τ)=fpower [v′ wind (t+τ)]

[0057] In the formula, P′ wind (t+τ) represents the predicted value of wind power generation; v′ wind (t+τ) represents the predicted wind speed; f power Represents the power curve function of the wind turbine.

[0058] Furthermore, based on the results of load forecasting and power generation forecasting, the supply and demand balance value in the future period is calculated, and the difference between power generation and load demand is determined according to the supply and demand balance value. According to the difference between power generation and load demand, a charging and discharging strategy of the energy storage unit is formulated, including:

[0059] Calculate the supply and demand balance value P in the future period balance (t+τ):

[0060]

[0061] In the formula, P′ PV (t+τ) represents the predicted photovoltaic power generation; P′ wind (t+τ) represents the predicted value of wind power generation; P′ load (t+τ) represents the load value predicted by simulation;

[0062] When P balance When (t+τ)>0, it means that the power generation is greater than the load demand, and the system needs to store the excess power in the energy storage unit; when P balance When (t+τ)<0, it means that the load demand is greater than the power generation, and the system needs to release electric energy from the energy storage unit to meet the load demand.

[0063] Furthermore, according to the current power state of the energy storage unit and the charging and discharging strategy, the power state of the energy storage unit is dynamically adjusted to ensure that the power is within the allowable range, including:

[0064] The charging and discharging power P of the energy storage unit battery (t) Dynamic adjustment, the calculation formula is as follows:

[0065]

[0066] Where η charg Represents the charging efficiency of the energy storage unit; η discharge Represents the discharge efficiency of the energy storage unit; P balance (t+τ) represents the supply and demand balance value in the future period;

[0067] The charge state of the energy storage unit is determined by the charge and discharge power P of the energy storage unit. battery (t) is updated, and its calculation formula is:

[0068] E battery (t+△t)=E battery (t)+P battery (t)·△t

[0069] 0≤E vattery (t)≤C battery (t)

[0070] In the formula, E battery (t) represents the energy storage unit charge at time t; Δt represents the time interval used to update the energy storage unit charge status; E battery (t+Δt) represents the energy storage unit charge state at the next moment t+Δt; C battery (t) represents the capacity of the energy storage unit.

[0071] Beneficial effects: Compared with the prior art, the significant technical effects of the present invention are:

[0072] The present invention uses a long short-term memory network (LSTM) to dynamically predict load demand. Compared with the fixed threshold scheduling method in the prior art, it can more accurately capture the complex fluctuation characteristics of the service area load and predict the load demand in the next 2 to 4 hours in advance. This prediction mechanism improves the system's response speed to load changes and reduces the energy waste and power shortage caused by prediction errors.

[0073] The present invention also combines weather data and time series analysis models to accurately predict photovoltaic power generation through a multivariate regression model, or accurately predict wind power generation through a wind speed time series model, to ensure that the energy storage system is charged when power generation is in excess and discharged when power generation is insufficient, thereby optimizing the charging and discharging strategy.

[0074] Compared with the traditional fixed-time charging and discharging mode, the present invention can flexibly adjust the energy storage operation according to the real-time supply and demand balance and the power grid electricity price signal, obtain electricity from the power grid for charging when the electricity price is low, and give priority to using the energy storage unit for power supply when the electricity price is high, which significantly reduces the operating cost of energy management and further improves the utilization efficiency of renewable energy. Through this method, the present invention realizes more efficient service area energy management, with obvious technical advantages and economic benefits.

[0075] The present invention is suitable for distributed energy scheduling and supply-demand balance management in service areas, and is particularly suitable for energy storage scheduling of intermittent renewable energy systems such as photovoltaic power generation and wind power generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a flow chart of a method for energy storage scheduling based on service area load and renewable energy generation prediction disclosed in an embodiment of the present invention;

[0077] Figure 2 It is a flow chart of a load forecasting method disclosed in an embodiment of the present invention;

[0078] Figure 3 It is a flow chart of an energy storage unit and a service area power dispatching strategy disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0079] The technical solution of the present invention is described in detail below in conjunction with specific implementation methods and the accompanying drawings of the specification.

[0080] See also Figure 1 , Figure 1 1 is a flow chart of a method for energy storage scheduling based on service area load and renewable energy generation prediction disclosed in an embodiment of the present invention. Figure 1 The energy storage scheduling method described is applied to power systems, such as distributed energy scheduling and supply and demand balance management in service areas, and the embodiments of the present invention do not limit this. Figure 1 As shown, the energy storage scheduling method based on service area load and renewable energy generation forecast may include the following operations:

[0081] S1. Data collection: Real-time collection of power generation, historical load data, and external characteristic data of power generation entities in the service area, where the external characteristic data includes weather data, traffic flow data, and time characteristics.

[0082] During this implementation, relevant data such as photovoltaic (wind power, etc.) power generation, power load, and meteorology in the service area are collected in real time.

[0083] S2. Load forecasting: Based on the historical load data and external characteristic data collected in real time, the long short-term memory network (LSTM) model is used to predict the load demand in the service area in the future period.

[0084] Based on historical load data, weather data (temperature, humidity, solar irradiance) and traffic flow data, a long short-term memory network (LSTM) is used to predict the load demand of the service area in the next 1-4 hours. The calculation steps include: data preprocessing, model training, load forecasting, result output and evaluation.

[0085] like Figure 2 As shown, the specific process of step S2 is as follows:

[0086] S2.1. Data preprocessing: Preprocess historical load data and external characteristic data to ensure data quality and consistency.

[0087] Preprocess historical load data, weather data, traffic data, time characteristics, etc. to ensure data quality and consistency. The data includes historical load data Pload (tn): represents the load demand at the past n time points; weather data (temperature, humidity, solar irradiance, etc.) W(tn): represents the climate conditions at the past n time points; service area traffic flow data T trafffc (tn): represents the traffic flow situation of the service area at the past n time points.

[0088] The data preprocessing steps include: filling or deleting missing values ​​in the data set; standardizing the load data and weather data so that they have the same numerical range to facilitate model processing; dividing the continuous historical load data into multiple time windows, each window containing a period of historical data, which is used to predict the load at the next moment.

[0089] S2.2, Model training: Use the pre-processed historical load data and external feature data to train the pre-built long short-term memory network model. During the training process, the model learns the dynamic relationship in the time series through historical load data and external feature data to optimize the prediction accuracy.

[0090] The training process of the model mainly includes: data input, forward propagation, loss function calculation, back propagation and weight update. In this process, the model learns the dynamic relationship in the time series through historical load data and external features to optimize the prediction accuracy. The specific process of step S2.2 is as follows:

[0091] S2.2.1. Data input: During the model training phase, data is input into the pre-built long short-term memory network model in the form of a time series window, which includes historical load data and external feature data.

[0092] During the model training phase, the input data is a time series window, including historical load data and external features. These input data will be input into the LSTM model as a multidimensional vector, and each time step contains load data and external features. Assuming the window size is n, the input of the model can be expressed as:

[0093] X={P load (tn), P load (t-n+1), ..., P load (t-1), P load (t)}

[0094] And the external features:

[0095] X external ={Tem(tn), Hum(tn), Solar(tn), T traffic (tn), ...}

[0096] Where Tem(tn) represents the temperature at the past n time points; Hum(tn) represents the humidity at the past n time points; Solar(tn) represents the solar irradiance at the past n time points.

[0097] S2.2.2.2, Forward propagation: Forward propagation is calculated progressively by multiple time steps (time series windows), and the output h of each time step is t Depends on the current input x t and the hidden state h at the previous moment t-1 ; The memory cell in the long short-term memory network model is the core of LSTM and is used to store information. It is passed along the entire sequence and can selectively save or discard information through a gating mechanism. This enables LSTM to preserve useful load data features for a long time while avoiding interference from irrelevant information. The following is the calculation process of the memory cell:

[0098] Forget gate calculation: Forget gate f i Control the information to be forgotten at the current moment to ensure that the network does not excessively remember irrelevant information; filter out historical load data that is irrelevant to future predictions through the forget gate:

[0099] f t =σ(W f ·[h t-1 , x t ]+b f )

[0100] Where j represents the activation function used to control the input, forget and output gates; W f represents the weight matrix of the forget gate; h t-1 represents the hidden state of the previous time step; x t Indicates the current input; b f Represents the bias term of the forget gate.

[0101] Input gate calculation: input gate i t It is used to selectively update the current input information to the memory unit. The input gate i t Determines how much of the current input information can be updated to the memory unit. Input gate i t Allows capturing new load variation patterns by selectively updating the memory state:

[0102] i t =σ(W i ·[h t-1 , x t ]+b i )

[0103] Where W i represents the weight matrix of the input gate; b irepresents the bias term of the input gate;

[0104] Candidate memory cell C′ t renew:

[0105] C′ i =tan h(W C ·[h t-1 , x t ]+b C )

[0106] In the formula, C′ t represents the candidate new cell state; W C represents the weight matrix of cell update; b C Represents the bias term for cell updating; tanh represents the hyperbolic tangent function, which is usually used to limit the range of output values ​​to (-1, 1);

[0107] Memory cell update: This step generates new candidate cell states to update the contents of the main memory cell. The formula is as follows:

[0108] C t =f t ·C t-1 +i t ·C′ t

[0109] In the formula, C t represents the cell state at the current time t; C t-1 Indicates the cell state at the previous moment;

[0110] Output gate calculation: output gate o t Control the cell state C at the current time t t What information needs to be output? The formula is:

[0111] o t =σ(W o ·[h t-1 , x t ]+b o )

[0112] In the formula, o t Represents the output gate value at the current moment, controlling the influence of the current cell state on the output at the next moment; W o represents the weight matrix of the output gate; b o Represents the bias term of the output gate.

[0113] The final hidden state h t Calculated based on the output gate and memory unit:

[0114] h t =o t ·tan h(Ct )

[0115] In the formula, h t Represents the hidden state at the current moment, that is, the output of the LSTM unit at the current moment;

[0116] Through multiple layers of calculation (i.e., recursive operation of time steps), hidden states are generated for each time step, which are used to predict future loads.

[0117] S2.2.3. Loss function calculation: The goal of model training is to minimize the error between the predicted value and the true value, and the error between the predicted value and the true value is calculated using the loss function.

[0118] In this embodiment, the loss function generally adopts mean square error (MSE), which is calculated as follows:

[0119]

[0120] In the formula, L(θ) represents the loss function; θ represents the model parameters; N represents the number of training samples; t represents the current time; τ represents the predicted time step; P′ load (t+τ) represents the load value predicted by simulation; P load (t+τ) represents the actual load value.

[0121] S2.2.4, Back propagation and weight update: After forward propagation and calculation of loss function, the error gradient is fed back to the parameters of each layer through the back propagation algorithm to adjust the weight matrix W i , W f , W C , W o and the bias term b i , b f , b C , b o , to minimize the loss function. Among them, the back propagation algorithm uses the gradient descent method for optimization.

[0122] In this embodiment, the back propagation algorithm is optimized using the gradient descent method, and the update formula is:

[0123]

[0124] In the formula, W represents the weight matrix; η represents the learning rate, which controls the step size of each update; Represents the gradient of the loss function with respect to the weights.

[0125] S2.3. Load forecasting: After the model training is completed, the latest historical load data and external feature data are input into the long short-term memory network model. The long short-term memory network model generates the load forecast value of the service area in the future period through recursive calculation.

[0126] In this embodiment, after the model training is completed, the LSTM model can perform load forecasting based on new input data. The forecasting process includes: inputting the latest historical load data and external characteristic data (weather, time, traffic, etc.), and the LSTM model generates a load forecast value for a future period (e.g., 1-4 hours) through recursive calculation.

[0127] P′ load (t+τ)=f LSTM [P load (tn), W(tn), T traffi c(tn)]

[0128] In the formula, t represents the current time; τ represents the predicted time step; P′ load (t+τ) represents the load value predicted by simulation; P load (tn) represents the electricity load demand at the past n time points; W(tn) represents the climate conditions (temperature, humidity, solar irradiance, etc.) at the past n time points; T trafffc (tn) represents the traffic flow in the service area at the past n time points.

[0129] S2.4. Result output and evaluation: Output the load forecast value of the service area in the future period predicted by the long short-term memory network model, and compare it with the actual value to evaluate the prediction accuracy.

[0130] If the error is large, the prediction effect can be further improved by adjusting model parameters, adding input features, or optimizing data preprocessing steps.

[0131] S3. Power generation forecast: Based on weather data, use the multivariate linear regression model to predict the photovoltaic power generation in the future period, or use the wind speed time series model to predict the future wind power generation; among them, the predicted photovoltaic power generation or wind power generation is determined according to the power generation entity.

[0132] If the service area is built with a photovoltaic power generation system, the output of photovoltaic power generation is closely related to the light intensity and temperature. PV (tn) and the current weather data W(t), the future power generation can be predicted by the multivariate linear regression model:

[0133] P′ PV (t+τ)=α 0 +α 1 ·S(t)+α 2 ·T(t)

[0134] In the formula, P′ PV (t+τ) represents the predicted photovoltaic power generation; α 0, α 1 , α 2 represents the regression coefficient obtained through training data; S(t) represents solar irradiance; T(t) represents temperature.

[0135] If the service area uses wind energy as clean energy, the prediction of wind power generation mainly depends on wind speed. Assume that the historical wind speed data is v wind (tn), predict the future wind speed through the wind speed time series model, and combine the power curve of the wind turbine to predict the future wind power generation:

[0136] P′ wina (t+τ)=f power [v′ wind (t+τ)]

[0137] In the formula, P′ wind (t+τ) represents the predicted value of wind power generation; v′ wind (t+τ) represents the predicted wind speed; f power Represents the power curve function of a wind turbine.

[0138] S4. Generation of energy storage dispatch strategy: Figure 3 As shown, based on the results of load forecasting and power generation forecasting, the supply and demand balance value in the future period is calculated, the difference between power generation and load demand is judged according to the supply and demand balance value, and the charging and discharging strategy of the energy storage unit is formulated according to the difference between power generation and load demand; according to the current power state of the energy storage unit and the charging and discharging strategy, the power state of the energy storage unit is dynamically adjusted to ensure that the power is within the allowable range; when the power of the energy storage unit reaches the upper or lower limit, according to the electricity price signal of the external power grid, it is decided whether to interact with the power grid, including purchasing electricity from the power grid or feeding electricity to the power grid, so as to form a final energy storage scheduling strategy to optimize energy utilization and economic benefits.

[0139] As an energy storage device, energy storage can balance the power generation and supply in the service area and maximize the economic benefits of energy through energy scheduling. This includes scheduling strategies for energy storage equipment and electricity consumption and power generation, as well as scheduling strategies for energy storage equipment and the power grid.

[0140] The specific process of step S4 is as follows:

[0141] S4.1. Based on the results of load forecast and power generation forecast, calculate the supply and demand balance value in the future period, determine the difference between power generation and load demand based on the supply and demand balance value, and formulate the charging and discharging strategy of the energy storage unit based on the difference between power generation and load demand.

[0142] The core of the charging and discharging strategy is to ensure the balance between power generation and load demand. First, calculate the supply and demand balance value P in the future period balance (t+τ):

[0143]

[0144] In the formula, P′ PV (t+τ) represents the predicted photovoltaic power generation; P′ wind (t+τ) represents the predicted value of wind power generation; P′ load (t+τ) represents the load value predicted by simulation;

[0145] When P balance When (t+τ)>0, it means that the power generation is greater than the load demand, and the system needs to store the excess power in the energy storage unit; when P balance When (t+τ)<0, it means that the load demand is greater than the power generation, and the system needs to release electric energy from the energy storage unit to meet the load demand. This is the basic strategy for charging and discharging the energy storage unit.

[0146] S4.2. Dynamically adjust the power state of the energy storage unit according to the current power state and charging and discharging strategy of the energy storage unit to ensure that the power is within the allowable range.

[0147] The charging and discharging power P of the energy storage unit battery (t) Dynamic adjustment, the calculation formula is as follows:

[0148]

[0149] Where η charge Represents the charging efficiency of the energy storage unit; η discharge Represents the discharge efficiency of the energy storage unit; P balance (t+τ) represents the supply and demand balance value in the future period;

[0150] The charge state of the energy storage unit is determined by the charge and discharge power P of the energy storage unit. battery (t) is updated, and its calculation formula is:

[0151] E battery (t+△t)=E battery (t)+P battery (t)·△t

[0152] 0≤E battery (t)≤C battery (t)

[0153] In the formula, E battery (t) represents the power of the energy storage unit at time t, indicating the power currently stored in the energy storage unit; Δt represents the time interval, which is used to update the power status of the energy storage unit; E battery (t+Δt) represents the energy storage unit charge state at the next moment t+Δt; C battery (t) represents the capacity of the energy storage unit.

[0154] S4.3. When the power of the energy storage unit reaches the upper or lower limit, it decides whether to interact with the grid based on the electricity price signal of the external grid, including purchasing electricity from the grid or feeding electricity to the grid, to form a final energy storage scheduling strategy to optimize energy utilization and economic benefits.

[0155] When the energy storage unit reaches the upper or lower limit, it decides whether to interact with the grid based on the electricity price signal of the external grid. The strategy of interacting with the external grid is to optimize the operating efficiency of the energy storage system through reasonable grid interaction when the power generation is insufficient or excessive, ensure the stability of energy supply in the service area, and reduce energy costs. Real-time monitoring of the electricity price signal of the grid P grid (t) As a reference for interaction with the power grid. The electricity price signal changes with the grid load fluctuation, and the system optimizes the charging and discharging decision based on the electricity price signal. Based on the current supply and demand balance and the grid electricity price, the energy storage unit makes the following charging and discharging decisions:

[0156] When the power generation is greater than the load, that is, the supply-demand balance is positive, part of the power generated is used for service area operations, and the remaining power is stored in the energy storage unit. When the energy storage unit reaches its maximum capacity, the excess power is sold to the grid;

[0157] When the load is greater than the power generation, that is, the supply and demand balance is negative, the electricity price data is introduced. When the electricity price is lower than the set threshold, the power grid is used to supply the power demand of the service area first, and the energy storage unit is charged for standby when the energy storage unit capacity is not full; when the electricity price is higher than the set threshold P low When the energy storage is insufficient, the power grid is used to supply the power demand in the service area.

[0158] The technical solution of the present invention predicts the load demand of the service area based on the long short-term memory network (LSTM), and combines the renewable energy power generation prediction (using the multivariate linear regression model to predict the photovoltaic power generation power in the future time period or using the wind speed time series model to predict the future wind power generation power) with the power grid electricity price signal to dynamically adjust the charging and discharging strategy of the energy storage unit. It is particularly suitable for service area scenarios with large fluctuations in load demand and unstable renewable energy power generation. By predicting the load fluctuations and renewable energy power generation in the service area, the present invention can automatically generate the optimal energy storage scheduling plan at different time periods, which is particularly suitable for scenarios with large traffic flow fluctuations and complex electricity demand in the service area. This method effectively reduces the service area's dependence on the external power grid and improves the stability and economic benefits of system operation.

[0159] The energy storage scheduling method based on service area load and renewable energy generation prediction of the present invention is verified by a specific embodiment as follows.

[0160] The energy storage scheduling method based on service area load and renewable energy generation prediction of the present invention includes:

[0161] S1. Take a service area as an example. The service area has built a photovoltaic project and deployed energy storage equipment. The load data, weather data and traffic flow data of the service area for nearly a year are collected as experimental samples to predict the power load and photovoltaic power generation.

[0162] S2. Power load forecast:

[0163] The experimental data include: (1) Historical load data. The service area power load (kW) per hour, the data range is from January 1, 2023 to December 31, 2023, a total of 8760 data points; (2) External characteristic data. Temperature: hourly temperature data (℃), humidity: hourly humidity data (%), solar irradiance: hourly solar irradiance data (W / m 2 ), Traffic flow: the number of vehicles passing through the service area per hour. Some sample data are shown in Table 1:

[0164] Table 1 Load forecast data sample

[0165]

[0166] Data preprocessing: Combine load data, temperature, humidity, solar irradiance, and traffic flow data into multi-dimensional input features on an hourly basis. Standardize load data and external features to ensure consistent data input ranges. Slide historical data by time window, using the past 24 hours of data as input to predict load demand for the next hour.

[0167] Model training: The input data is the load and external feature data of the past 24 hours, and the output data is the load demand for the next hour. The data from January 2023 to October 2023 is used as the training set, and the data from November 2023 to December 2023 is used as the validation set. After model training, the model parameters are set as follows: number of hidden layer units: 50, time step: 24, learning rate: 0.001.

[0168] The data from December 31, 2023 was used for testing. The load, temperature, humidity and other characteristics of the past 24 hours were input to predict the load demand for the next 24 hours on January 1, 2024. The prediction results are shown in Table 2:

[0169] Table 2 Load forecast results (partial)

[0170] time Actual load (kW) Forecast load (kW) 2024-01-01 00:00 125 123 2024-01-01 01:00 118 119 2024-01-01 02:00 110 112 2024-01-01 03:00 105 108 2024-01-01 04:00 102 104 2024-01-01 05:00 98 99 … … …

[0171] The average prediction error of the model is 2.3%. In the prediction results, the model can accurately capture the changing trend of load, especially during peak hours at night and during the day. The prediction results are highly consistent with the actual load.

[0172] S3. Photovoltaic power generation forecast:

[0173] The experimental data include: (1) Photovoltaic power generation data. Photovoltaic power generation power (kW) per hour, the data range is from January 1, 2023 to December 31, 2023, a total of 8760 data points; (2) External characteristic data. Temperature: hourly temperature data (℃), solar irradiance: hourly solar irradiance data (W / m2). Some sample data are shown in Table 3:

[0174] Table 3 Photovoltaic power generation data sample

[0175] time Photovoltaic power generation (kW) Temperature(℃) <![CDATA[Solar irradiance (W / m 2 )]]> … … … … 2023-06-01 06:00 0 20 0 2023-06-01 07:00 10 21 150 2023-06-01 08:00 35 23 320 2023-06-01 09:00 60 25 600 2023-06-01 10:00 80 27 720 2023-06-01 11:00 95 28 820 … … … …

[0176] Data preprocessing: Before conducting PV power generation forecasting, data needs to be preprocessed to ensure that the model can accurately identify patterns in the data, including missing value processing, standardization, and time window sliding.

[0177] Model training: In the photovoltaic power generation prediction, a multivariate linear regression model is used for modeling. The model can automatically learn the influence of solar irradiance and temperature on photovoltaic power generation through training data. The data from January to October 2023 is used as the training set (accounting for 80% of the total data) to train the regression model. The data from November to December 2023 is used as the test set (accounting for 20%) to verify the prediction performance of the model. Regression coefficient after training: α 0 =5.2, the coefficient of solar irradiance α 1 =0.35, temperature coefficient α 2 =1.8.

[0178] Power generation forecast: Using the trained multivariate regression model, the photovoltaic power generation on January 1, 2024 is predicted. The results are shown in Table 4:

[0179] Table 4 Photovoltaic power generation prediction results (partial)

[0180]

[0181]

[0182] The average error is 3.2%, and the model's prediction performance is good, especially during periods of high solar irradiance, when the predicted values ​​are highly consistent with the actual values.

[0183] S4. Energy storage dispatch strategy:

[0184] Taking the data of part of the time period on January 1, 2024 as an example, the balance of power generation and load supply and demand is calculated as shown in Table 5:

[0185] Table 5 Supply and demand balance calculation results (partial)

[0186] time Forecast load (kW) Forecast photovoltaic power generation (kW) Supply and demand balance (kW) … … … … 2024-01-01 06:00 100 0 -100 2024-01-01 07:00 110 12 -98 2024-01-01 08:00 125 38 -87 2024-01-01 09:00 140 65 -75 2024-01-01 10:00 155 175 +20 2024-01-01 11:00 170 190 +20 2024-01-01 12:00 180 185 +5 2024-01-01 13:00 185 160 -25 … … … …

[0187] The dispatch strategy of energy storage units in the service area is as follows:

[0188] When the power generation is greater than the load, that is, the supply-demand balance is positive, part of the power generated is used for service area operations, and the remaining power is stored in the energy storage unit. When the energy storage unit reaches its maximum capacity, the excess power is sold to the grid;

[0189] When the load is greater than the power generation, that is, the supply and demand balance is negative, the electricity price data is introduced. When the electricity price is lower than the set threshold, the power grid is used first to supply the power demand of the service area, and the energy storage unit is charged for standby when the energy storage unit capacity is not full; when the electricity price is higher than the set threshold, energy storage is used first to supply power, and when the energy storage capacity is insufficient, the power grid is used to supply the power demand of the service area.

Claims

1. A method for energy storage scheduling based on service area load and renewable energy generation forecast, characterized in that: include: Data collection: Real-time collection of power generation, historical load data, and external characteristic data of power generation entities in the service area, where the external characteristic data includes weather data, traffic flow data, and time characteristics; Load forecasting: Based on the historical load data and external characteristic data collected in real time, the long short-term memory network model is used to predict the load demand in the service area in the future period; Power generation forecast: Based on weather data, use the multivariate linear regression model to predict the photovoltaic power generation in the future period, or use the wind speed time series model to predict the future wind power generation; the photovoltaic power generation or wind power generation is determined by the power generation entity; Energy storage dispatch strategy generation: Based on the results of load forecast and power generation forecast, calculate the supply and demand balance value in the future period, judge the difference between power generation and load demand based on the supply and demand balance value, and formulate the charging and discharging strategy of the energy storage unit based on the difference between power generation and load demand; dynamically adjust the power state of the energy storage unit according to the current power state of the energy storage unit and the charging and discharging strategy to ensure that the power is within the allowable range; when the power of the energy storage unit reaches the upper or lower limit, decide whether to interact with the power grid, including purchasing electricity from the power grid or feeding electricity to the power grid, according to the electricity price signal of the external power grid, to form the final energy storage dispatch strategy to optimize energy utilization and economic benefits.

2. The energy storage scheduling method based on service area load and renewable energy generation prediction according to claim 1 is characterized in that: Based on the historical load data and external characteristic data collected in real time, the long short-term memory network model is used to predict the load demand in the service area in the future period, including: Data preprocessing: preprocess historical load data and external characteristic data to ensure data quality and consistency; Model training: Use the pre-processed historical load data and external feature data to train the pre-built long short-term memory network model. During the training process, the model learns the dynamic relationship in the time series through the historical load data and external feature data to optimize the prediction accuracy; Load forecasting: After model training is completed, the latest historical load data and external feature data are input into the long short-term memory network model. The long short-term memory network model generates the load forecast value of the service area in the future period through recursive calculation; Result output and evaluation: Output the load forecast value of the service area in the future period predicted by the long short-term memory network model, and compare it with the actual value to evaluate the prediction accuracy.

3. The energy storage scheduling method based on service area load and renewable energy generation prediction according to claim 2 is characterized in that: The pre-processed historical load data and external feature data are used to train the pre-built long short-term memory network model. During the training process, the model learns the dynamic relationship in the time series through the historical load data and external feature data to optimize the prediction accuracy, including: Data input: During the model training phase, data is input to the pre-built LSTM network model in the form of a time series window, which includes historical load data and external feature data; Forward propagation: Forward propagation is calculated progressively by multiple time series windows, and the output h of each time series window is t Depends on the current input x t and the hidden state h at the previous moment t-1 ; The memory units in the LSTM network model are used to store information. The memory units are transmitted along the entire sequence, and the information is selectively saved or discarded through the gating mechanism, so that the LSTM network model can store useful load data features for a long time while avoiding interference from irrelevant information; Loss function calculation: The goal of model training is to minimize the error between the predicted value and the true value. The error between the predicted value and the true value is calculated using the loss function. Back propagation and weight update: After forward propagation and calculation of the loss function, the error gradient is fed back to the parameters of each layer through the back propagation algorithm, and the weight matrix and bias term are adjusted to minimize the loss function; among them, the back propagation algorithm uses the gradient descent method for optimization.

4. The energy storage scheduling method based on service area load and renewable energy generation prediction according to claim 3 is characterized in that: The forward propagation includes: Forget gate calculation: Forget gate f t Control the information to be forgotten at the current moment to ensure that the network does not excessively remember irrelevant information; filter out historical load data that is irrelevant to future predictions through the forget gate: f t =σ(W f ·[h t-1 ,x t ]+b f ) Where σ represents the activation function used to control the input, forget and output gates; W f represents the weight matrix of the forget gate; h t-1 represents the hidden state of the previous time step; x t Indicates the current input; b f Represents the bias term of the forget gate; Input gate calculation: input gate i t It is used to selectively update the current input information to the memory unit. The input gate i t Allows capturing new load variation patterns by selectively updating the memory state: i t =σ(W i ·[h t-1 ,x t ]+b i ) Where W i represents the weight matrix of the input gate; b i represents the bias term of the input gate; Candidate memory cell C t ′ renew: C t ′ =tanh(W C ·[h t-1 ,x t ]+b C ) In the formula, C t ′ represents the candidate new cell state; W C represents the weight matrix of cell update; b C Represents the bias term for cell update; tanh represents the hyperbolic tangent function, which is usually used to limit the range of output values ​​to (-1,1); Memory cell update: This step generates new candidate cell states to update the contents of the main memory cell. The formula is as follows: C t =f t ·C t-1 +i t ·C′ t In the formula, C t represents the cell state at the current time t; C t-1 Indicates the cell state at the previous moment; Output gate calculation: output gate o t Control the cell state C at the current time t t The information that needs to be output is as follows: the t =σ(W o ·[h t-1 ,x t ]+b o ) In the formula, o t Represents the output gate value at the current moment, controlling the influence of the current cell state on the output at the next moment; W o represents the weight matrix of the output gate; b o Represents the bias term of the output gate; The final hidden state h t Calculated based on the output gate and memory unit: h t =o t ·tanh(C t ) In the formula, h t Represents the hidden state at the current moment, that is, the output of the LSTM unit at the current moment; Through multiple layers of calculation, the hidden state of each time step is generated, and all hidden states are used to predict future loads.

5. The energy storage scheduling method based on service area load and renewable energy generation prediction according to claim 3 is characterized in that: The goal of model training is to minimize the error between the predicted value and the true value. The error between the predicted value and the true value is calculated using the loss function, including: The loss function uses mean square error, which is calculated as follows: Where L(θ) represents the loss function; θ represents the model parameters; N represents the number of training samples; t represents the current time; τ represents the predicted time step; P l ′ oad (t+τ) represents the load value predicted by simulation; P load (t+τ) represents the actual load value.

6. The energy storage scheduling method based on service area load and renewable energy generation prediction according to claim 3 is characterized in that: The back propagation algorithm uses the gradient descent method for optimization, and the update formula is: In the formula, W represents the weight matrix; η represents the learning rate, which controls the step size of each update; Represents the gradient of the loss function with respect to the weights.

7. The energy storage scheduling method based on service area load and renewable energy generation prediction according to claim 2 is characterized in that: After the model training is completed, the latest historical load data and external feature data are input into the long short-term memory network model. The long short-term memory network model generates the load forecast value of the service area in the future period through recursive calculation, including: P′ load (t+τ)=f LSTM [P load (t-n),W(t-n),T traffic (t-n)] In the formula, t represents the current time; τ represents the predicted time step; P′ load (t+τ) represents the load value predicted by simulation; P load (tn) represents the electricity load demand at the past n time points; W(tn) represents the climate conditions at the past n time points; T traffic (tn) represents the traffic flow in the service area at the past n time points.

8. The energy storage scheduling method based on service area load and renewable energy generation prediction according to claim 1, characterized in that: Based on weather data, the photovoltaic power generation in the future period is predicted using a multivariate linear regression model, or the wind power generation in the future is predicted using a wind speed time series model; the prediction of photovoltaic power generation or wind power generation is determined based on the power generation entity, including: If the service area uses a photovoltaic power generation system, the output of photovoltaic power generation is closely related to the light intensity and temperature. Based on the historical power generation data P PV (tn) and current weather data W(t), and predict future power generation through a multivariate linear regression model: P′ PV (t+τ)=α0+α1·S(t)+α2·T(t) In the formula, P′ PV (t+τ) represents the predicted photovoltaic power generation; α0, α1, α2 represent the regression coefficients obtained through training data; S(t) represents solar irradiance; T(t) represents temperature; If the service area uses wind energy as clean energy, the prediction of wind power generation depends on wind speed. It is assumed that the historical wind speed data is v wind (tn), predict the future wind speed through the wind speed time series model, and combine the power curve of the wind turbine to predict the future wind power generation: P′ wind (t+τ)=f power [v′ wind (t+τ)] In the formula, P′ wind (t+τ) represents the predicted value of wind power generation; v′ wind (t+τ) represents the predicted wind speed; f power Represents the power curve function of the wind turbine.

9. The energy storage scheduling method based on service area load and renewable energy generation prediction according to claim 1, characterized in that: Based on the results of load forecast and power generation forecast, the supply and demand balance value in the future period is calculated, and the difference between power generation and load demand is judged according to the supply and demand balance value. According to the difference between power generation and load demand, the charging and discharging strategy of the energy storage unit is formulated, including: Calculate the supply and demand balance value P in the future period balance (t+τ): Where P P ′ V (t+τ) represents the predicted photovoltaic power generation power; P ′ wind (t+τ) represents the predicted value of wind power generation; P i ′ oad (t+τ) represents the load value predicted by simulation; When P balance When (t+τ)>0, it means that the power generation is greater than the load demand, and the system needs to store the excess power in the energy storage unit; when P balance When (t+τ)<0, it means that the load demand is greater than the power generation, and the system needs to release electric energy from the energy storage unit to meet the load demand.

10. The energy storage scheduling method based on service area load and renewable energy generation prediction according to claim 1, characterized in that: According to the current power state and charging and discharging strategy of the energy storage unit, the power state of the energy storage unit is dynamically adjusted to ensure that the power is within the allowable range, including: The charging and discharging power P of the energy storage unit battery (t) Dynamic adjustment, the calculation formula is as follows: Where η char Represents the charging efficiency of the energy storage unit; η discharge Represents the discharge efficiency of the energy storage unit; P balance (t+τ) represents the supply and demand balance value in the future period; The charge state of the energy storage unit is determined by the charge and discharge power P of the energy storage unit. battery (t) is updated, and its calculation formula is: E battery (t+Δt)=E battery (t)+P battery (t)·Δt 0≤E battery (t)≤C battery (t) In the formula, E battery (t) represents the energy storage unit charge at time t; Δt represents the time interval used to update the energy storage unit charge status; E battery (t+Δt) represents the energy storage unit charge state at the next moment t+Δt; C battery (t) represents the capacity of the energy storage unit.

Images