Multi-energy system double-layer optimization scheduling method for prediction error dynamic compensation

Abstract

The invention relates to a prediction error dynamic compensation multi-energy system double-layer optimization scheduling method. The method comprises the following steps: S1, performing day-ahead wind power and photovoltaic power prediction according to historical wind and light data; s2, respectively constructing prediction error distribution of wind power output and photovoltaic output; s3, constructing an upper-layer optimization scheduling model including wind and light output and an energy storage charging and discharging strategy, and based on dynamic compensation of the wind and light prediction error, aiming at reducing the wind and light abandoning amount, improving the consumption rate of renewable energy sources, and meanwhile, requiring to reduce net load fluctuation and determining a curve of residual load borne by a thermal power generating unit; s4, constructing a lower-layer optimization scheduling model of the output of the thermal power generating unit, and considering the economical efficiency of system operation by taking the minimum system operation cost as an optimization target; and S5, solving the double-layer model by using an optimization algorithm, and generating an optimal scheduling strategy of the multi-energy system.

Description

Technical Field

[0001] The present invention belongs to the field of optimal dispatching of power systems, and particularly relates to a two-layer optimal dispatching method for a multi-energy system with dynamic compensation of prediction errors. Background Art

[0002] With the accelerating transformation of the global energy structure towards low-carbon and clean energy, the installed capacity of renewable energy sources represented by wind power and photovoltaic power has been continuously climbing and has gradually become the main power source of the power system. However, the power generation characteristics of wind power and photovoltaic power have significant randomness and volatility, posing severe challenges to the safe and economic operation of the power system.

[0003] As a typical real-time balancing system, the operation mechanism of the power system requires that the power generation output and the load demand always maintain a dynamic balance. Traditional dispatching methods usually formulate plans based on deterministic prediction results. However, the prediction errors of wind power and photovoltaic power will cause the actual output of renewable energy to deviate from the dispatching plan, resulting in an imbalance between power supply and demand, the phenomenon of wind and light abandonment, forcing the dispatching strategy to reserve additional reserve capacity and frequently adjust the unit output, leading to increased costs and aggravated equipment wear. At the same time, the uncertainty of wind power and photovoltaic power output will increase the amplitude of net load fluctuations, increase the peak shaving pressure of thermal power units, and reduce the overall operation reliability and economy of the system. To achieve the large-scale and efficient utilization of renewable energy sources such as wind and light power generation and ensure the economic and safe operation of the power system, it is necessary to establish a dynamic response dispatching model considering uncertainty for wind and light power prediction deviations to solve the problems of insufficient system regulation ability and increased operation risks in the scenario of high-proportion renewable energy grid connection. Summary of the Invention

[0004] The purpose of the present invention is to provide a two-layer optimal dispatching method for a multi-energy system with dynamic compensation of prediction errors, which can improve the consumption rate of renewable energy while reducing the system operation cost and peak shaving pressure, and provide reliable technical support for a high-proportion renewable energy power system.

[0005] To achieve the above objectives, the present invention includes the following steps:

[0006] S1. Respectively perform day-ahead power predictions of wind power and photovoltaic power according to historical wind and light data;

[0007] S2. Respectively construct the prediction error distributions of wind power output and photovoltaic power output;

[0008] S3. Construct an upper-layer optimal dispatching model including wind and light output and energy storage charge and discharge strategies, based on the dynamic compensation of the prediction errors, aiming to reduce the amount of wind and light abandonment, improve the consumption rate of renewable energy, and at the same time require reducing the net load fluctuations, and determine the remaining load curve borne by thermal power units;

[0009] S4. Build the lower-layer optimal scheduling model for the output of thermal power units, with the minimum system operation cost as the optimization goal, considering the economy of system operation;

[0010] S5. Use the optimization algorithm to solve the bi-level model and generate the optimal scheduling strategy for the multi-energy system.

[0011] In the wind power and photovoltaic power prediction stage, build a Transformer-LSTM prediction model. In the present invention, the Transformer decoding layer is modified into a fully connected layer, and an LSTM module is added. While simplifying the traditional Transformer model, its adaptability to time series prediction is enhanced.

[0012] Through the automatic learning ability of the deep neural network, deeply analyze the historical meteorological data and power data to realize the prediction of the daily power generation for the next 24 hours. Apply the predicted wind power output and photovoltaic power output data to the subsequent optimal scheduling.

[0013] Build the prediction error distribution of wind power output and photovoltaic power output. Build the prediction error distribution of photovoltaic power generation according to the daily cycle irradiance intensity distribution law, and at the same time use fuzzy random variables to build the wind power prediction error distribution. The specific method is as follows:

[0014] The prediction error of photovoltaic power output has a high correlation with the prediction error of solar radiation intensity. According to the daily cycle radiation intensity distribution law, the prediction error of radiation intensity can be expressed as a normal distribution variable with a mean of 0 and a standard deviation of σ k . Since the photovoltaic power output is linearly related to the irradiance intensity, according to the properties of the normal distribution, it can be obtained that the prediction errors of photovoltaic power output at different times still follow the normal distribution, which is The probability density function is:

[0015]

[0016] To characterize the uncertainty characteristics of wind power prediction, considering that the wind power prediction error has random volatility and ambiguity, the present invention uses the fuzzy random theory of uncertainty for modeling. The expression of the wind power prediction error at time t is:

[0017]

[0018] In the formula, represents the wind power output prediction error, and v w,t represents a random variable, q e is the deviation magnitude obtained according to historical statistical data. Decompose to get:

[0019]

[0020] The probability distribution of the wind-solar output prediction error obtained based on the above method is introduced into the scheduling model to construct an optimal scheduling framework for dynamic compensation of prediction deviation.

[0021] An upper-layer optimal scheduling model including wind-solar output and energy storage charge-discharge strategies is constructed. The optimization objectives are to minimize the curtailment of wind and solar power and the variance of the net load. The constraints include wind power output constraints, photovoltaic power output constraints, and energy storage system constraints. The decision variables are the predicted wind-solar power, load demand, and energy storage charge-discharge strategies, aiming to improve the consumption of wind and solar energy, reduce the peak-valley difference of the load, and relieve the peak-shaving pressure of thermal power units in the lower-layer model.

[0022] The objective function for the consumption of renewable energy with dynamic compensation of prediction error is:

[0023]

[0024] In the formula, △t is the duration of each time period, and in this paper, △t = 1h; represents the curtailed wind power of the wind farm at time t; is the curtailed photovoltaic power of the PV power station at time t, and the formula is as follows:

[0025]

[0026] In the formula, and are the predicted powers of wind power and photovoltaic power at time t respectively; p w,t and p v,t are the actual output powers of wind power and photovoltaic power at time t respectively.

[0027] The consumption level of system renewable energy is measured by counting the total amount of curtailed wind and solar power during the scheduling period. The prediction error distribution is used for dynamic adjustment of the consumption power, forming a dynamic compensation mechanism with the ability to adapt to prediction errors, effectively improving the consumption rate of renewable energy.

[0028] The objective function for net load fluctuation is:

[0029]

[0030] In the formula, P glt,t is the net load value at time t; P l,t is the load value at time t; is the discharge power of the energy storage system at time t; is the charge power of the energy storage system at time t; P glt,av is the average value of the net load within a scheduling period.

[0031] Taking the minimum net load variance as the optimization objective, the flexible charging and discharging capabilities of energy storage are utilized to follow the fluctuations of wind power, photovoltaic power, and load, effectively shaving peaks and filling valleys, obtaining the optimal wind power and photovoltaic power outputs and the energy storage charging and discharging powers for each period, and transferring the remaining load curve after deducting the wind power, photovoltaic power, and energy storage outputs to the lower-level model.

[0032] The wind power output constraint is:

[0033]

[0034] In the formula, is the maximum wind power output within the scheduling period.

[0035] The photovoltaic power output constraint is:

[0036]

[0037] In the formula, is the maximum photovoltaic power output within the scheduling period.

[0038] The energy constraint of the energy storage system is:

[0039] S min ≤S es,t ≤S max

[0040] In the formula, S es,t is the state of charge of the energy storage system at time t; δ represents the self-discharge rate of the energy storage; η c and η d represent the charging and discharging efficiencies of the energy storage system respectively; E s is the capacity of the energy storage system; S max and S min are the upper and lower limits of the state of charge of the energy storage system respectively.

[0041] The energy storage charging and discharging constraint is:

[0042]

[0043] In the formula, and are the charging and discharging state variables of the energy storage system, taking values of 0 or 1; and are the minimum and maximum charging powers respectively; and are the minimum and maximum discharging powers respectively.

[0044] Build the lower-layer optimal scheduling model for the output of thermal power units, which is characterized by correlating the remaining load curve transmitted by the upper-layer model, taking the output constraints, start-stop constraints, and ramping constraints of thermal power units as constraint conditions, and aiming at minimizing the operating cost of thermal power units to optimize the economic operation of thermal power units.

[0045] The economic objective function of thermal power units is:

[0046]

[0047] C co is the coal consumption cost of thermal power units, which is calculated using the consumption characteristics. a i , b i and c i are the consumption coefficients of thermal power unit i respectively; P i,t is the output of thermal power unit i at time t; C st is the start-stop cost of thermal power units; S i,t represents the start-up cost coefficient of thermal power unit i; u i,t is the start-stop state of thermal power unit i, taking 0 or 1.

[0048] The output constraint of thermal power units is:

[0049] u i,t p i,min ≤p i,t ≤u i,t p i,max

[0050] In the formula, P i,min and P i,max are the upper limit and lower limit of the output of thermal power unit i respectively.

[0051] The start-stop constraint of thermal power units is:

[0052]

[0053] In the formula, T on and T off are the maximum continuous on-time and maximum continuous off-time of thermal power unit i respectively.

[0054] The ramping constraint of thermal power units is:

[0055] -r i,down ≤p i,t -p i,(t-1) ≤r i,up

[0056] In the formula, r i,down and r i,up are the maximum downward ramping rate and maximum upward ramping rate of thermal power units respectively.

[0057] The power balance constraint is as follows:

[0058]

[0059] The particle swarm optimization algorithm is used to solve the bi-level model and generate the optimal scheduling strategy for the multi-energy system. The particle swarm optimization algorithm has a fast convergence speed and can efficiently solve the complex optimization problems in the bi-level model; its swarm intelligence characteristics can effectively coordinate the multi-objective constraint relationships and have wide applicability in the field of optimal power system scheduling.

[0060] Compared with the existing inventions, the beneficial effects achieved by the present invention are as follows:

[0061] By integrating the global feature extraction ability of Transformer and the time series modeling advantage of LSTM, a hybrid prediction framework is constructed to significantly improve the accuracy of wind and light power prediction and provide reliable input for subsequent error compensation.

[0062] Using the prediction error dynamic compensation mechanism, in the upper-layer optimization model, the probability distribution of wind and light prediction errors is introduced to construct a deviation adaptive compensation framework, which significantly improves the system's active absorption ability for prediction deviations, realizes the dynamic balance between the scheduling plan and the operating state, reduces the regulation frequency of conventional units, and effectively alleviates the impact of prediction uncertainty on system balance; at the same time, through the coordinated optimization of the charge and discharge strategy of energy storage systems and wind and light output, a scheduling boundary with smooth net load is generated; the lower-layer model conducts economic dispatch of thermal power units based on boundary conditions to form an upper-lower coordinated decision-making. A hierarchical iterative solution algorithm is designed for the structural characteristics of the bi-level model to enable the hierarchical optimization of the upper-layer prediction error compensation strategy and the lower-layer economic dispatch objective, avoiding the problem that it is difficult to solve due to too many variables in the traditional single model. Brief Description of the Drawings

[0063] Figure 1 It is a structural diagram of a bi-level optimal scheduling method for a multi-energy system with dynamic compensation of prediction errors provided by the present invention.

[0064] Figure 2 It is a Transformer-LSTM wind and light output prediction model constructed by the present invention.

[0065] Figure 3 It is a diagram of the bi-level optimal scheduling strategy for the multi-energy system implemented by the present invention. Detailed Embodiments

[0066] To make the technical solutions and their advantages of the present invention clearer, the typical implementation cases will be elaborated in detail below with reference to the drawings.

[0067] It should be noted that the listed embodiments are only partial implementation solutions of the present invention and do not constitute all limitations of the technical solutions.

[0068] As Figure 2 shown, the Transformer-LSTM wind and light output prediction model constructed by the present invention is as follows:

[0069] Although LSTM can retain certain long-term dependence information through the gating mechanism, as the number of time steps increases, the propagated gradient signal will gradually weaken, and important information may still be lost. The multi-head attention mechanism of Transformer can interact with other positions of the entire input sequence at each position through parallel computing and direct attention, so as to extract more distant temporal features. Therefore, the present invention modifies the decoding layer into a fully connected layer and adds an LSTM module, which simplifies the traditional Transformer model and makes it more suitable for time series prediction.

[0070] The Transformer module marks the positions of data through the parallel position encoding layer to solve the problem of missing position information caused by parallel computing. Its core multi-head attention mechanism realizes global dependence modeling through feature space mapping, effectively alleviating the gradient attenuation phenomenon of traditional recurrent neural networks in long sequence prediction. The feature vectors after residual connection then enter the feed-forward neural layer to implement non-linear transformation, mapping the features into a representation form suitable for processing by the LSTM network. The LSTM constructs a temporal processor, and realizes dynamic feature screening through the cell state gating system, retaining the local temporal pattern of the sequence data, and finally completes the temporal prediction of the power data through the fully connected layer.

[0071] As Figure 3 shown, the upper-layer optimization model of the multi-energy system with dynamic compensation of prediction error takes the wind power output constraint, photovoltaic power output constraint and energy storage system constraint as the constraint conditions, takes the wind and light predicted power, the distribution of wind and light output prediction error, the load demand and the energy storage charge and discharge strategy as the decision variables, and takes the renewable energy consumption and the net load fluctuation as the objective function.

[0072] In the specific implementation of the present invention, a wind farm with a total capacity of 300 MW and a photovoltaic power station with a total capacity of 50 MW are used, and the parameters of the energy storage system are shown in Table 1.

[0073]

[0074] The renewable energy consumption level of the system is measured by statistically counting the total amount of abandoned wind and light during the dispatching period. Based on the fuzzy random distribution of wind power and the daily cycle normal distribution of photovoltaic power, the present invention constructs a stochastic optimization objective function with confidence level constraints, and uses the prediction error distribution for the dynamic adjustment of the consumption power, forming a dynamic compensation mechanism with the ability to adapt to prediction errors, effectively improving the renewable energy consumption rate.

[0075] To relieve the peak shaving pressure of the thermal power units in the lower-layer model, the flexible charge and discharge capacity of energy storage is utilized to follow the fluctuations of wind power, photovoltaic power and load, effectively shaving peaks and filling valleys, obtaining the optimal wind power and photovoltaic power output and the charge and discharge power of energy storage at each time period, and transferring the remaining load curve after deducting the output of wind power, photovoltaic power and energy storage to the lower-layer model.

[0076] In the specific implementation of the present invention, 5 thermal power units are used, and the specific parameters are shown in Table 2.

[0077]

[0078] According to Figure 3 , the lower-layer optimal scheduling model of the multi-energy system with dynamic compensation for prediction error is associated with the remaining load curve transferred by the upper-layer model. With the output constraints, start-stop constraints and ramp constraints of the thermal power units as the constraint conditions and the minimization of the operating cost of the thermal power units as the objective, the economic operation of the thermal power units is optimized.

[0079] The above-listed implementation schemes are preferred examples, and their descriptions do not constitute a limitation on the implementation modes of the technical solutions. The scheme adjustments, equivalent replacements of technical solutions and technical improvement schemes implemented within the core essence scope of the present technical solution should all be included in the protection scope of the present invention.

Claims

1. A two-layer optimal scheduling method for a multi-energy system with dynamic compensation of prediction errors, characterized in that It includes the following steps: S1. Perform day-ahead power predictions for wind power and photovoltaic respectively based on historical scenery data; S2. Construct prediction error distributions for wind power output and photovoltaic output respectively; S3. Construct an upper-layer optimal scheduling model including scenery output and energy storage charge and discharge strategies, and perform dynamic compensation based on the scenery prediction error, aiming to reduce the curtailment of wind and light, improve the consumption rate of renewable energy, and at the same time require reducing the net load fluctuation, and determine the remaining load curve borne by thermal power units; S4. Construct a lower-layer optimal scheduling model for thermal power unit output, with the minimum system operation cost as the optimization goal, considering the economy of system operation; S5. Use an optimization algorithm to solve the two-layer model and generate the optimal scheduling strategy for the multi-energy system.

2. The power prediction of current wind power and photovoltaic according to claim 1, characterized in that Construct a Transformer-LSTM prediction model, modify the Transformer decoding layer into a fully connected layer, and add an LSTM module. Through the automatic learning ability of the deep neural network, deeply analyze the historical meteorological data and power data to realize the prediction of daily power generation power within the next 24 hours, and apply the predicted wind power output and photovoltaic output data to the subsequent optimal scheduling.

3. The prediction error distribution of wind power output and photovoltaic power output is constructed according to claim 1, characterized in that, Construct a photovoltaic power generation prediction error distribution according to the daily cycle irradiance intensity distribution law, and at the same time use fuzzy random variables to construct a wind power prediction error distribution. The specific method is as follows: The prediction error of photovoltaic output is highly correlated with the prediction error of solar radiation intensity. According to the daily cycle radiation intensity distribution law, the prediction error of radiation intensity can be expressed as a normal distribution variable with a mean of 0 and a standard deviation of σ k , and since the photovoltaic output is linearly related to the irradiation intensity, according to the properties of the normal distribution, it can be obtained that the prediction errors of photovoltaic output at different times still follow a normal distribution, which is The probability density function is: To characterize the uncertainty characteristics of wind power prediction, considering that the wind power prediction error has random volatility and fuzziness, the present invention uses the fuzzy random theory of uncertainty for modeling. The expression of the wind power prediction error at time t is: In the formula, represents the prediction error of wind power output, and v w,t represents a random variable, q e is the deviation magnitude obtained based on historical statistical data. Decompose to get:

4. The upper-layer optimal scheduling model for constructing the wind and light output and energy storage charge and discharge strategies according to claim 1, wherein With the minimum curtailment of wind and light and the minimum net load variance as the optimization goals, with wind power output constraints, photovoltaic output constraints, and energy storage system constraints as the constraint conditions, and with the predicted scenery power, scenery output prediction error distribution, load demand, and energy storage charge and discharge strategy as decision variables, aiming to improve the consumption of scenery, reduce the peak-valley difference of the load, and relieve the peak-shaving pressure of thermal power units in the lower-layer model; The objective function of renewable energy consumption with dynamic compensation of prediction error is: where Δt is the duration of each time period, and in this paper, Δt = 1 h; represents the curtailed wind power of the wind farm in time period t; is the curtailed photovoltaic power of the PV power plant in time period t, and the formula is as follows: In the formula, and are the predicted power of wind power and photovoltaic power in period t respectively; p w,t and p v,t are the actual output power of wind power and photovoltaic power in period t respectively; The system's renewable energy consumption level is measured by statistically counting the total curtailment of wind and light within the scheduling period. The prediction error distribution is used for the dynamic adjustment of the consumption power to form a dynamic compensation mechanism with the ability to adapt to prediction errors, effectively improving the consumption rate of renewable energy; The objective function of net load fluctuation is: Where, P glt,t is the net load value at time t; P l,t is the load value at time t; is the discharge power of the energy storage system at time t; is the charging power of the energy storage system at time t; P glt,av is the average value of the net load within a scheduling period; With the minimum net load variance as the optimization goal, use the flexible charge and discharge ability of the energy storage to follow the fluctuations of scenery and load, effectively shaving peaks and filling valleys, obtain the optimal scenery output and energy storage charge and discharge power at each time period, and transfer the remaining load curve after deducting the output of scenery, energy storage, and load to the lower-layer model; The wind power output constraint is: Wherein, is the maximum wind power output within the scheduling period; The photovoltaic output constraint is: In the formula, is the maximum value of the photovoltaic output during the scheduling period; The energy constraint of the energy storage system is: S min ≤ S es,t ≤ S max where S es,t is the state of charge of the energy storage system at time t; δ represents the self-discharge rate of the energy storage; η c and η d respectively represent the charging and discharging efficiencies of the energy storage system; E s is the energy storage system capacity; S max and S min are respectively the upper and lower limits of the state of charge of the energy storage system; The energy storage charge and discharge constraint is: Wherein, and are respectively the charge and discharge state variables of the energy storage system, taking values of 0 or 1; and are respectively the minimum and maximum charging powers; and are respectively the minimum and maximum discharging powers.

5. The lower-layer optimal scheduling model for constructing the output of a thermal power unit according to claim 1, wherein Associate with the remaining load curve transferred by the upper-layer model, with the thermal power unit output constraint, start-stop constraint, and ramp constraint as the constraint conditions, and with the minimum operation cost of the thermal power unit as the goal, optimize the economic operation of the thermal power unit; The economic objective function of the thermal power unit is: C co is the coal consumption cost of the thermal power unit, which is calculated using the consumption characteristics, a i , b i and c i are respectively the consumption coefficients of thermal power unit i; P i,t is the output of thermal power unit i at time t; C st is the start-stop cost of the thermal power unit; S i,t represents the start-up cost coefficient of thermal power unit i; u i,t It is the start-stop state of thermal power unit i, taking 0 or 1; The thermal power unit output constraint is: u i,t p i,min ≤p i,t ≤u i,t p i,max Wherein, P i,min and P i,max are respectively the upper limit and the lower limit of the output of thermal power unit i; The thermal power unit start-stop constraint is: where, T on and T off are respectively the maximum continuous starting time and the maximum continuous shutdown time of thermal power unit i; The thermal power unit ramp constraint is: -r i,down ≤ p i,t -p i,(t-1) ≤ r i,up where r i,down and r i,up are the maximum downward ramp rate and the maximum upward ramp rate of the thermal power unit, respectively; The power balance constraint is:

6. A two-layer optimal scheduling method for a multi-energy system with dynamic compensation of prediction errors according to claim 1, characterized in that, The particle swarm optimization algorithm is used to solve the bi-level model and generate the optimal scheduling strategy for the multi-energy system. The particle swarm optimization algorithm has a fast convergence speed and can efficiently solve complex optimization problems in the bi-level model; Its swarm intelligence characteristics can effectively coordinate multi-objective constraint relationships and have wide applicability in the field of optimal power system scheduling.

Images