Wind power prediction error interval evaluation method based on system scheduling reserve classification

Abstract

The invention discloses a wind power prediction error interval evaluation method based on system scheduling standby classification, and belongs to the field of wind power prediction and comprehensive energy system optimization scheduling. Wind power historical data is used for participating in system scheduling, and wind power prediction errors are classified according to a standby power range obtained by system scheduling; historical wind power data and historical prediction errors are used as model input, a prediction error interval is used as output, an error interval evaluation model is established, the influence of prediction step length and wind power waveforms on wind power prediction error types is analyzed, and the scheduling cycle of the system is dynamically adjusted on the basis of the obtained prediction error types. According to the method, wind power prediction and system scheduling are combined, the accuracy and reference of wind power interval evaluation are improved, the scheduling period of the system is guided and adjusted on this basis, and the stability of the system and the reliability of scheduling are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction and integrated energy system optimal dispatching, and particularly relates to a method for evaluating the error interval of wind power prediction based on the classification of system dispatching reserve. Background Art

[0002] New energy has gradually become the main energy source, but the high uncertainty of new energy has seriously threatened the reliability and security of power system operation. Taking wind power as an example, the current research on coping with the uncertainty of wind power mainly focuses on two aspects: prediction and dispatching. At the prediction level, either an advanced prediction model is used for deterministic prediction, or the uncertainty of wind power prediction is analyzed, that is, the interval evaluation of wind power.

[0003] Currently, more new models for deterministic prediction mainly consider the correlation between numerical weather prediction (NWP) and wind power, the spatio-temporal characteristics of wind energy, or prediction models considering extreme weather. However, no matter what kind of wind power modeling technology is adopted, if only the point prediction value of wind power is given, it cannot fully reflect the uncertainty of wind power. As an extension of point prediction, interval evaluation provides the range within which the point prediction result probably deviates from the actual power. It is widely regarded as a necessary condition for optimal power system management because it quantifies the uncertainty associated with point prediction. However, the existing interval classification methods are all divided according to the characteristics of the prediction model or wind power, ignoring the impact of the coupling between system dispatching and wind power prediction on the prediction result.

[0004] At the dispatching level, domestic and foreign scholars improve the dispatching flexibility of the system and the ability to cope with the uncertainty of wind power by considering the energy storage capacity of new energy forms, studying the characteristics of equipment with energy storage potential in the system for the consumption effect of wind power, or proposing dispatching strategies considering the uncertainty of wind power. However, no matter which method is used, the wind power prediction error will affect the reliability of the integrated energy system dispatching. The greater the error, the lower the accuracy of the dispatching plan. If the predicted value is too large, load shedding is likely to occur, affecting the safe and stable operation of the system. If the predicted value is too small, a large amount of wind abandonment is likely to occur, affecting the economic operation of the system. Therefore, in power system dispatching, the rolling dispatching method is usually adopted to reduce the adverse impact of wind power prediction error on the system by shortening the dispatching cycle. This actually utilizes the general law that the shorter the prediction step, the higher the prediction accuracy.

[0005] Whether it is the scheduling within 4 hours, 2 hours or 1 hour, or the real-time scheduling of 15 minutes, they are all artificially fixed scheduling cycles, which are formulated based on the common 1-step, 4-step, 8-step and 16-step prediction results in wind power prediction. As the proportion of wind power gradually increases, using a fixed scheduling cycle for rolling scheduling can no longer guarantee the reliability of the system. If the scheduling instruction release cycle is too long, the large multi-step prediction error of wind power will lead to a significant imbalance between power generation and consumption; if the scheduling instruction release cycle is too short, the unit output adjustment will be too frequent, bringing potential risks to the stable operation of the system. In the future when the proportion of wind power is getting higher and higher, dynamically adjusting the system scheduling cycle is a powerful means to cope with the uncertainty of wind power, but it is necessary to provide the wind power prediction error corresponding to different numbers of steps in the scheduling cycle. In addition, in the case of sudden rise and fall of wind power, the wind power prediction error is likely to change sharply in a short time. At this time, if the system still uses a fixed scheduling cycle to formulate the scheduling plan, a large amount of wind curtailment or load shedding is likely to occur, affecting the stability of the system. Therefore, it is necessary to study the influence of the prediction step length and wind power waveform on the wind power prediction error and explore new ideas for wind power interval prediction.

[0006] In addition, in the existing research on coping with the uncertainty of wind power, prediction and scheduling either exist in isolation, or the prediction only provides the original wind power or wind power prediction error data for scheduling, and the main body still focuses on the scheduling strategy itself. But in fact, in order to achieve a more reasonable scheduling effect, the two can be more closely combined. Using the historical scheduling results to provide theoretical and data support for wind power interval prediction, the obtained wind power prediction interval can more reasonably guide the formulation of future scheduling plans. Summary of the Invention

[0007] The present invention provides a method for evaluating the wind power prediction error interval based on the classification of system scheduling reserves, which combines wind power prediction with system scheduling, improves the accuracy and referenceability of wind power interval evaluation, and guides the adjustment of the system scheduling cycle based on this, improving the stability of the system and the reliability of scheduling.

[0008] An embodiment of the present invention provides a method for evaluating the wind power prediction error interval based on the classification of system scheduling reserves, including the following steps:

[0009] Step 1, establish a comprehensive energy system available reserve model based on historical wind power;

[0010] Step 2, establish a wind power prediction error classification model based on the available reserves of the system;

[0011] Step 3: Based on the available reserve model of the integrated energy system and the wind power prediction error classification model, establish a wind power prediction error interval evaluation model based on the system dispatching reserve classification. Use the wind power prediction error interval evaluation model to predict the type of wind power prediction error, and adjust the dispatching period of the integrated energy system according to the predicted type of wind power prediction error.

[0012] Optionally, in an embodiment of the present invention, step 1 specifically includes the following steps:

[0013] Step 1.1: Establish the cost objective function C of the integrated energy system and solve it respectively:

[0014] minC = min(C CON + C CHP + C QF )

[0015] In the formula: C is the total system cost, C CON is the operating cost of the thermal power unit, C CHP is the operating cost of the combined heat and power unit, C QF is the penalty cost for wind curtailment;

[0016]

[0017] In the formula: N1 is the number of thermal power units, N2 is the number of combined heat and power units, t total is the total dispatching period, a k , b k , c k are the operating cost coefficients of thermal power unit k, α m , β m , γ m , δ m , θ m , μ m are the operating cost coefficients of heat and power unit m, c 1 is the price of steam coal, ζ QF is the wind curtailment penalty coefficient, P CON,k,t is the electrical output of thermal power unit k at time t, P CHP,m,t is the electrical output of combined heat and power unit m at time t, Q CHP,m,t is the heat output of CHP unit m at time t, P wind,t is the wind power output at time t, are the upper and lower limits of the wind power output at time t respectively;

[0018] Step 1.2: Solve the constraint conditions satisfied when solving the cost objective function C:

[0019] The system power balance constraint is divided into electrical power balance and heat power balance, and the electrical power balance and heat power balance constraints are as follows respectively:

[0020]

[0021] Where: P LD,t is the total electrical load power at time t, P n,t is the charge / discharge power of the energy storage device, P n,t >0 indicates energy storage, P n,t <0 indicates discharging, Q LD,t is the total heat load power at time t;

[0022] The unit operation constraints mainly include unit output constraints, unit ramp constraints, and thermoelectric coupling constraints of cogeneration units. Among them, the output constraints of thermal power units and cogeneration units are as follows:

[0023]

[0024] Where: P CON is the electric power of the thermal power unit, are the upper and lower limits of the electric power of the thermal power unit respectively; P CHP is the electric power of the cogeneration unit, are the upper and lower limits of the electric power of the cogeneration unit respectively, Q CHP is the thermal power of the cogeneration unit, are the upper and lower limits of the thermal power of the cogeneration unit respectively, P wind is the wind power output, are the upper and lower limits of the wind power output respectively;

[0025] The ramp constraints of thermal power units and cogeneration units are as follows:

[0026]

[0027] Where: are the sliding power and ramp power of the thermal power unit respectively, are the sliding power and ramp power of the cogeneration unit respectively;

[0028] The cogeneration unit has thermoelectric coupling constraints, and the expression is as follows:

[0029]

[0030] Where: C m is the thermoelectric ratio of cogeneration unit m under back pressure condition, c m is the thermoelectric ratio of cogeneration unit m under maximum condensing condition, K m is a constant;

[0031] The constraints of the electricity storage device are divided into two categories. One is the constraint on the charge-discharge command X, and the other is the constraint on the charge-discharge power and the capacity of the electricity storage device:

[0032]

[0033] In the formula: is the charge command and discharge command of the electricity storage device at time t, is the charge power and discharge power of the electricity storage device, is the maximum charge-discharge power and minimum charge-discharge power that can be provided within a single scheduling period, E n,t 、 are the electricity storage capacity, maximum available capacity and minimum available capacity of the electricity storage device at time t, η chu and η f are the charging efficiency and discharging efficiency of the electricity storage device;

[0034] Step 1.3: Solve the available reserve model of the integrated energy system:

[0035] The available reserve of the integrated energy system includes the total reserve power provided by the conventional units and the reserve power provided by the electricity storage device after the scheduling ends, which are divided into storable reserve power and dischargeable reserve power, and are respectively expressed as:

[0036]

[0037] In the formula: are respectively the total storable reserve power and total dischargeable reserve power that the system can provide at time t, are respectively the storable reserve power and dischargeable reserve power provided by the thermal power unit at time t, is the storable reserve power and dischargeable reserve power provided by the combined heat and power unit at time t, are respectively the storable reserve power and dischargeable reserve power provided by the electricity storage device at time t, is the maximum charge-discharge power that the electricity storage device can provide.

[0038] Optionally, in an embodiment of the present invention, step 2 specifically includes:

[0039] Considering the correlation between the system's storable / dischargeable reserve power curve and the positive and negative prediction errors of wind power, the wind power prediction errors are divided into 5 categories:

[0040] When the wind power prediction error is positive and greater than the maximum value R of the system's dischargeable reserve power a,fang,max at this time, the current wind power prediction error is set as type I;

[0041] When the wind power prediction error is positive and between the maximum value and the minimum value of the system's dischargeable reserve power, the current wind power prediction error is set as type II;

[0042] The wind power prediction error is positive and less than the minimum value R of the system's available reserve power a,fang,min , or the wind power prediction error is negative and its absolute value is less than the minimum value R of the system's reserve power a,chu,min At this time, the current wind power prediction error is set as Class III;

[0043] When the wind power prediction error is negative and its absolute value is between the maximum and minimum values of the system's reserve power, the current wind power prediction error is set as Class IV;

[0044] The wind power prediction error is negative and its absolute value is greater than the maximum value R of the system's reserve power a,chu,max At this time, the current wind power prediction error is set as Class V.

[0045] Optionally, in an embodiment of the present invention, step 3 specifically includes the following steps:

[0046] Step 3.1: Use the LSTM neural network to establish a wind power prediction error interval evaluation model based on the system dispatch reserve classification;

[0047] The LSTM neural network includes an input layer, a recursive hidden layer, and an output layer. The reading and modification of the memory unit in the LSTM are realized by controlling the input gate, forget gate, and output gate. The specific calculation formula is as follows:

[0048]

[0049] In the formula: W xc , W xi , W xf , W xo are the weight matrices connecting the input signal x t , W hc , W hi , W hf , W ho are the weight matrices connecting the output signal h t of the hidden layer, W ci , W cf , W co is the diagonal matrix connecting the output vector c t of the neuron activation function and the gate function, b i , b c , b f , b o are the bias vectors, and σ is the activation function;

[0050] Taking the wind power data and the wind power prediction error as the input, and the future wind power prediction error type and interval as the output, use the correlation between the input and output of the LSTM neural network to establish a wind power prediction error interval evaluation model. The expression is:

[0051] h t+1 = f(x t , x t-1 , …, x t-n )

[0052] In the formula: h t+1 is the type of prediction error, and x i is the input data, which is composed of wind power data P i and wind power prediction error data e i , and x i = (P i , e i ), where i = t, t - 1, …, t - n;

[0053] Step 3.2: Solve the evaluation indexes of the wind power prediction error interval evaluation model:

[0054] Use three evaluation indexes, namely the accuracy of the evaluation error category, the coverage rate of the evaluation interval, and the average bandwidth of the interval, to verify the reliability of the wind power prediction error interval evaluation model:

[0055] (1) The accuracy of the evaluation error category A CC :

[0056]

[0057] In the formula: N is the number of evaluation samples. When the original error category is the same as the evaluation error category, k i is 1, otherwise it is 0; The value range of A CC is [0, 1], representing the accuracy of the evaluation model. The closer A CC is to 1, the higher the evaluation accuracy; The smaller A CC is, the lower the accuracy and the worse the interval evaluation effect;

[0058] (2) The coverage rate of the evaluation interval I PICP :

[0059]

[0060] In the formula: If the evaluation target value falls within the evaluation interval, c i is 1, otherwise it is 0; The greater the frequency that the actual wind power value falls within the evaluation interval, that is, the closer the interval coverage rate is to 1, the better the interval prediction effect;

[0061] (3) The average bandwidth of the interval I PINAW :

[0062] The interval width δ of the evaluation interval is the difference between the upper bound U and the lower bound L of the interval, that is:

[0063] δ = U - L

[0064] The interval average bandwidth is as follows:

[0065]

[0066] Where: R is the difference between the maximum and minimum values of the test sample, and I PINAW The smaller it is, the narrower the prediction interval;

[0067] Step 3.3: According to the obtained type of prediction error interval, determine whether the integrated energy system needs to adjust the scheduling period at the current moment.

[0068] Optionally, in an embodiment of the present invention, the method for adjusting the scheduling period of the integrated energy system according to the predicted wind power prediction error type is as follows:

[0069] If within the next 4 hours from the current moment, the wind power prediction errors at each time point corresponding to the prediction steps are all of type III, directly formulate the scheduling plan for the next 4 hours at the current moment;

[0070] If within the next 4 hours from the current moment, only some discontinuous time points have wind power prediction errors of type II or type IV corresponding to the prediction steps, still formulate the scheduling plan for the next 4 hours at the current moment;

[0071] If within the preset time period in the future from the current moment, the wind power prediction errors at each time point corresponding to the prediction steps are all of type III, and the wind power prediction errors at each time point corresponding to the prediction steps within the next 2h - 4h are all of type II or type IV, adjust the scheduling period of the integrated energy system at the current moment and formulate the scheduling plan for the next 2 hours;

[0072] If within the next 4 hours from the current moment, the prediction error at a certain time point corresponding to the prediction step is of type I or type V, issue a new scheduling instruction before the current moment and enter a new scheduling period.

[0073] The method for evaluating the wind power prediction error interval based on system scheduling reserve classification in the embodiment of the present invention makes the wind power interval prediction result more convenient for determining the scheduling period of the system and better serves the system scheduling. First, use the historical wind power data to participate in the system scheduling, and guide the classification of future wind power prediction errors according to the reserve power range provided by the system historical scheduling. Then, use the historical wind power data and historical prediction errors as the model inputs and the future prediction error type as the output, and establish an error interval evaluation model using the LSTM neural network. Finally, consider the influence of the prediction step and the wind power waveform on the prediction error type, verify with domestic and foreign wind power data as an example, and dynamically adjust the scheduling period of the system using the obtained prediction error type. The method proposed in the present invention further combines the wind power prediction with the system scheduling, improves the accuracy and referenceability of the wind power interval evaluation, and guides the adjustment of the system scheduling period based on this, improving the stability of the system and the reliability of the scheduling.

[0074] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. Description of the Drawings

[0075] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of embodiments in conjunction with the drawings, wherein:

[0076] Figure 1 It is a flowchart of a method for evaluating the error interval of wind power prediction based on system dispatch reserve classification according to an embodiment of the present invention;

[0077] Figure 2 It is a schematic diagram of the method for evaluating the error interval of wind power prediction based on system dispatch reserve classification according to an embodiment of the present invention;

[0078] Figures 3(a) - 3(d) It is a graph of the evaluation results of the wind power prediction interval under different numbers of steps (1 step, 4 steps, 12 steps, and 16 steps) of Elia data according to an embodiment of the present invention;

[0079] Figure 4 It is a graph of the classification results of the prediction error of 1-16 steps under Elia data according to an embodiment of the present invention;

[0080] Figures 5(a) and 5(b) are graphs of the evaluation index results of three models under different wind power ratios (wind power ratio of 11% and wind power ratio of 20%) of Elia data according to an embodiment of the present invention;

[0081] Figures 6(a), 6(b), and 6(c) are graphs of the system dispatch cycle adjustment results under different wind power ratios (wind power ratio of 11%, wind power ratio of 16%, and wind power ratio of 20%) of Elia data according to an embodiment of the present invention;

[0082] Figure 7 It is a graph of the classification results of the prediction error of 1-16 steps of onshore wind farm data in China according to an embodiment of the present invention;

[0083] Figure 8 It is a graph of the system dispatch cycle adjustment result of onshore wind farm data in China according to an embodiment of the present invention. Detailed Embodiments

[0084] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0085] Generally speaking, in wind power prediction, the larger the prediction step, the larger the wind power prediction error and error range. In system scheduling, the longer the scheduling period, the lower the scheduling reliability at the end moment of the scheduling period. As the proportion of wind power increases year by year, the existing scheduling strategies with fixed scheduling periods cannot guarantee the reliability of the system. However, the dynamic adjustment of the system scheduling period should not only ensure the stability of the system but also avoid frequent adjustments of the unit output, and it needs to match the wind power prediction errors with different prediction steps. Therefore, in order to make the wind power prediction results more convenient for determining the system scheduling period, reduce the adverse effects brought by wind power uncertainty, and thus improve the reliability of system scheduling, the present invention proposes a method for evaluating the wind power prediction error range based on system scheduling reserve. First, the historical wind power data is used to participate in system scheduling, and the wind power prediction errors are classified according to the reserve power range provided by the system. Then, an evaluation model for the wind power prediction error range is established, with the historical wind power data and historical wind power prediction errors as the model inputs and the future wind power prediction error types and ranges as the outputs, using the correlation between the inputs and outputs of the LSTM neural network deep learning model. Finally, the obtained wind power prediction error types are used to dynamically adjust the system scheduling period. In the numerical example, the effectiveness of the proposed method for evaluating the wind power prediction error range of the present invention is verified using domestic and foreign wind power data. Compared with the existing wind power prediction error classification methods based on prediction model features or wind power features, the method proposed in the present invention further integrates wind power prediction with system scheduling, uses the historical scheduling reserve power of the system to guide the classification of future wind power prediction errors, and at the same time explores the influence of the prediction step and wind power waveform on the wind power prediction error types, providing a basis for adjusting the system scheduling period.

[0086] Figure 1 FIG. is a flowchart of a method for evaluating the wind power prediction error range based on system scheduling reserve classification according to an embodiment of the present invention.

[0087] As Figure 1 shown, the method for evaluating the wind power prediction error range based on system scheduling reserve classification includes the following steps:

[0088] Step 1, establish an available reserve model of the integrated energy system based on historical wind power.

[0089] Step 1.1: Establish an integrated energy system cost objective function C and solve it respectively:

[0090] minC = min(C CON + C CHP + C QF )

[0091] In the formula: C is the total system cost; C CON is the operating cost of the thermal power unit, C CHP is the operating cost of the combined heat and power (CHP) unit; CQF is the curtailment penalty cost. Specifically, there are:

[0092]

[0093] In the formula: N1 is the number of thermal power units; N2 is the number of combined heat and power (CHP) units; t total is the total scheduling period; a k , b k , c k are the operating cost coefficients of thermal power unit k; α m , β m , γ m , δ m , θ m , μ m are the operating cost coefficients of CHP unit m; c 1 is the price of thermal coal; ζ QF is the curtailment penalty coefficient; P CON,k,t is the electrical output of thermal power unit k at time t; P CHP,m,t is the electrical output of CHP unit m at time t; Q CHP,m,t is the heat output of CHP unit m at time t; P wind,t is the wind power output at time t; are the upper and lower limits of the wind power output at time t, respectively.

[0094] Step 1.2: When solving the objective function C, a series of constraint conditions need to be satisfied:

[0095] The system power balance constraint is divided into electrical power balance and heat power balance. The electrical power balance and heat power balance constraints are as follows, respectively:

[0096]

[0097] In the formula: P LD,t is the total electrical load power at time t; P n,t is the charge / discharge power of the energy storage device. When P n,t > 0, it means energy storage, and when P n,t < 0, it means discharging; Q LD,t is the total heat load power at time t.

[0098] The unit operation constraints mainly include unit output constraints, unit ramp rate constraints, and thermoelectric coupling constraints of CHP units. The output constraints of thermal power units and CHP units are as follows:

[0099]

[0100] In the formula: P CON is the electrical power of the thermal power unit, are its upper and lower limits, respectively; P CHPis the electric power of the CHP unit, and its upper and lower limits; Q CHP is the thermal power of the CHP unit, and its upper and lower limits; P wind is the wind power output, and its upper and lower limits.

[0101] The ramp rate constraints of the thermal power unit and the CHP unit are as follows:

[0102]

[0103] In the formula: are the sliding and ramp rates of the thermal power unit respectively; are the sliding and ramp rates of the CHP unit respectively.

[0104] In addition, the CHP unit also has a thermoelectric coupling constraint, and its expression is as follows:

[0105]

[0106] In the formula: C m is the thermoelectric ratio of the cogeneration unit m under the back pressure condition; c m is the thermoelectric ratio of the CHP unit m under the maximum condensing condition; K m is a constant.

[0107] The constraint conditions of the energy storage device are mainly divided into two categories. One is the constraint on the charge-discharge command X, and the other is the constraint on the charge-discharge power and the capacity of the energy storage device:

[0108]

[0109] In the formula: is the charge-discharge command of the energy storage device at time t; is the charge-discharge power of the energy storage device; are the maximum and minimum charge-discharge powers that can be provided within a single scheduling period; E n,t 、 are the stored energy capacity, the maximum and minimum available capacities of the energy storage device at time t; η chu and η f are the charge-discharge efficiencies of the energy storage device.

[0110] Step 1.3: Solve the dispatchable reserve model:

[0111] The available reserve of the integrated energy system includes the total reserve power provided by the conventional units and a part of the reserve power provided by the energy storage device after the dispatch ends. It can be divided into storable reserve power and dischargeable reserve power, which can be specifically expressed as:

[0112]

[0113] In the formula: are the total storable / dischargeable reserve power that the system can provide at time t, respectively; are the storable / dischargeable reserve power provided by thermal power units at time t, respectively; is the storable / dischargeable reserve power provided by the CHP unit at time t; are the storable / dischargeable reserve power provided by the energy storage device at time t, respectively, is the maximum charge / discharge power that the energy storage device can provide.

[0114] Step 2: Establish a wind power prediction error classification model based on the available reserve of the system.

[0115] Step 2.1: Considering the correlation between the storable / dischargeable reserve power curve of the system and the positive and negative prediction errors of wind power, the wind power prediction errors are divided into 5 categories. The specific classification method is as Figure 2 , and the idea is as follows:

[0116] When the wind power prediction error (in this invention, the wind power prediction error = wind power predicted value - wind power actual value) is positive and greater than the maximum value R of the system's dischargeable reserve power a,fang,max , it indicates that if the system dispatching plan is set based on such a wind power prediction result, the system cannot fully compensate for the wind power prediction error, resulting in a large amount of load shedding. Therefore, such a wind power prediction error is set as Class I;

[0117] When the wind power prediction error is positive and between the maximum and minimum values of the system's dischargeable reserve power, it indicates that at this time the system can compensate for most of the wind power prediction errors, resulting in a small amount of load shedding. Other compensation measures of the system can be further considered. Therefore, such a wind power prediction error is set as Class II;

[0118] When the wind power prediction error is positive and less than the minimum value R of the system's dischargeable reserve power a,fang,min , or the wind power prediction error is negative and its absolute value is less than the minimum value R of the system's storable reserve power a,chu,min , it indicates that at this time the system can fully compensate for the wind power prediction error at any time. Therefore, such a wind power prediction error is set as Class III;

[0119] When the wind power prediction error is negative and its absolute value is between the maximum and minimum values of the system's storable reserve power, analogous to Class II prediction errors, it indicates that at this time the system only generates a small amount of wind curtailment. Therefore, such a wind power prediction error is set as Class IV;

[0120] When the wind power prediction error is negative and its absolute value is greater than the maximum value R of the system's storable reserve power a,chu,max , analogous to Class I prediction errors, at this time the system cannot fully compensate for the wind power prediction error, resulting in a large amount of wind curtailment. Therefore, such a wind power prediction error is set as Class V.

[0121] Step 3: Based on the available reserve model of the integrated energy system and the wind power prediction error classification model, establish a wind power prediction error interval evaluation model based on system dispatching reserve classification. Use the wind power prediction error interval evaluation model to predict the type of wind power prediction error, and adjust the dispatching cycle of the integrated energy system according to the predicted type of wind power prediction error.

[0122] Step 3.1: Use the LSTM neural network to establish a wind power prediction error interval evaluation model based on system dispatching reserve classification:

[0123] The LSTM artificial neural network is a neural network for processing and predicting sequential data. It can learn the long-term dependence information of the input data to prevent the vanishing gradient and exploding gradient of information transmission, thereby enhancing its ability to capture the dynamic changes of time series. Therefore, LSTM is also widely used in wind speed and wind power prediction. The LSTM neural network consists of three parts: an input layer, a recurrent hidden layer, and an output layer. The reading and modification of the memory unit in the LSTM are realized by controlling the input gate, forget gate, and output gate. The specific calculation formulas are as follows:

[0124]

[0125] In the formula: W xc 、W xi 、W xf 、W xo are the weight matrices connecting the input signal x t ; W hc 、W hi 、W hf 、W ho are the weight matrices connecting the output signal h t of the hidden layer; W ci 、W cf 、W co are the diagonal matrices connecting the output vector c t of the neuron activation function and the gate function; b i 、b c 、b f 、b o are the bias vectors; σ is the activation function, usually the tanh or sigmoid function.

[0126] For the input data, if only power is considered, the influence of different evaluation models on the evaluation of error categories is ignored; if only errors are considered, the original fluctuations of the wind power time series are ignored. Therefore, a data set composed of the errors obtained from the point prediction model and the wind power data is used as the input of the evaluation model, so that the LSTM network can not only learn the original fluctuations of the wind power, but also learn the different error distributions brought by different point prediction methods. The output data is changed to the error category, that is, the evaluation model can directly obtain the error category from the power and the error, and the expression of the evaluation model is:

[0127] h t+1 = f(x t , x t-1 , …, x t-n )

[0128] where: h t+1 is the predicted error type; x i is the input data, which is composed of the wind power data P i and the predicted error data e i , that is, x i = (P i , e i ), i = t, t - 1, …, t - n.

[0129] Step 3.2: Solve the model evaluation index:

[0130] Three types of evaluation indexes, namely evaluation accuracy rate, evaluation interval coverage rate, and interval average bandwidth, are used to verify the reliability of the error evaluation model.

[0131] (1) Evaluation error category accuracy rate A CC :

[0132]

[0133] where: N is the number of evaluation samples; when the original error category and the evaluated error category are the same, k i is 1, otherwise it is 0; the value range of A CC is [0, 1], and its value represents the accuracy of the evaluation model. The closer A CC is to 1, the higher the evaluation accuracy; the smaller A CC is, the lower the accuracy and the worse the interval evaluation effect.

[0134] (2) Evaluation interval coverage rate I PICP :

[0135]

[0136] where: if the evaluation target value falls within the evaluation interval, c iIt is 1; otherwise it is 0. The greater the frequency at which the actual wind power value falls within the evaluation interval, that is, the closer the interval coverage rate is to 1, the better the interval prediction effect or the higher the reliability.

[0137] (3) Interval average bandwidth I PINAW :

[0138] The interval width δ of the evaluation interval is the difference between the upper bound U and the lower bound L of the interval, that is:

[0139] δ = U - L

[0140] The interval average bandwidth is:

[0141]

[0142] In the formula: R represents the difference between the maximum value and the minimum value of the test sample. I PINAW The smaller it is, the narrower the prediction interval and the better the model sensitivity.

[0143] Step 3.3: According to the type of prediction error interval obtained by the above method, determine whether the system needs to adjust the scheduling period at the current moment. Combining Figure 2 , with 15 minutes as one step, the specific method for adjusting the system scheduling period is as follows:

[0144] If within the next 4 hours from the current moment, the wind power prediction errors at each time point corresponding to the prediction steps are all of type III, it means that the system can fully compensate for the wind power prediction errors using its own standby power at any moment within the next 4 hours from the current moment, and the scheduling plan for the next 4 hours can be directly formulated at the current moment;

[0145] If within the next 4 hours from the current moment, only the wind power prediction errors at some discontinuous time points corresponding to the prediction steps are of type II or type IV, it means that its prediction errors are close to the critical point of type III. At this time, the system only generates a small amount of wind curtailment or load shedding. To avoid frequent adjustment of the unit, the scheduling plan for the next 4 hours is still formulated at the current moment;

[0146] If within a certain period (assumed to be 2 hours) within the next 4 hours from the current moment, the wind power prediction errors at each time point corresponding to the prediction steps are all of type III, while the wind power prediction errors at each time point corresponding to the prediction steps within 2h - 4h in the future are all of type II or type IV, it means that formulating the scheduling plan for the next 4 hours at the current moment will result in a large amount of load shedding or wind curtailment. Therefore, the system scheduling period is adjusted at the current moment, and the scheduling plan for the next 2 hours is formulated;

[0147] If within the next 4 hours from the current moment, the prediction error at a certain time point corresponding to the prediction step is of type I or type V, it means that the system cannot compensate for the prediction error at this moment at all, affecting the stability of the system. At this time, a new scheduling instruction is issued before this moment, and a new scheduling period starts from this moment.

[0148] The following uses a specific embodiment to illustrate the method for evaluating the wind power prediction error interval based on system dispatching reserve classification of the present invention.

[0149] As Figure 2 shown, in the example system of the present invention, the source side includes a thermal power plant equipped with two pure condensing thermal power units; a cogeneration power plant equipped with two cogeneration units; a wind farm, the output range of the units is shown in Table 1, and the rated power of the energy storage device is 50 MW. The dispatching time interval of the integrated energy system dispatching model based on historical wind power in the present invention is set to 15 minutes, and the Gurobi solver is called to solve it in the MATLAB environment.

[0150] Table 1 Output range of each unit

[0151] Unit Electric output range / MW Thermal output range / MW CHP1 [75,150] [0,200] CHP2 [75,150] [0,200] CON1 [50,100] / CON2 [50,100] /

[0152] The original wind power data adopted in the present invention is sourced from the operation data publicly released by the Elia Belgian power operator and the measured power data provided by a certain wind farm in China. Among them, the data on the Elia website is selected for the time period from October to December 2023, and the data of the domestic wind farm is for the time period from January to February 2019, with a time interval of 15 minutes for both. 2000 sample points are taken as the pre-training set, 2000 sample points as the validation set, 1000 sample points as the training set, and 1000 sample points as the test set. The LSTM network model part is established based on the deep learning framework, using Python as the programming language, with the compilation environment being PyCharm Community Edition 2024, the processor being 12th Gen Intel(R) Core(TM) i5-12500H, and the RAM being 16GB.

[0153] Simulation results:

[0154] Analysis of interval evaluation results:

[0155] The wind power prediction error categories classified according to the system dispatching reserve power are combined with the wind power and wind power prediction error data in the original time series order to form a new data set, which is substituted into the LSTM neural network interval evaluation model for training to obtain the interval evaluation results for predictions 1 - 16 steps in advance. Take the interval evaluation results for several different steps (1 step, 4 steps, 12 steps, 16 steps) as Figures 3(a) - 3(d) shown. It can be Figures 3(a) - 3(d) seen that the method proposed in the present invention has a certain effect on the interval evaluation of different steps, and overall shows a trend that the smaller the number of steps, the better the effect.

[0156] Three evaluation indicators are calculated based on the evaluation results to evaluate the wind power interval prediction effect of the method of the present invention. In addition, different wind power ratios will cause changes in the range of reserve power obtained after its participation in system dispatching, which will in turn lead to different classification results and affect the wind power prediction effect. Specifically, the increase in the proportion of wind power output will inevitably lead to a decrease in the output of conventional units. Correspondingly, the available reserve that can be provided by conventional units decreases, and the available release reserve increases. This results in a worse tolerance of the system to negative wind power prediction errors and an increase in the compensation ability for positive prediction errors. Therefore, in order to further explore the influence of the wind power ratio on the strategy of the present invention, interval predictions are respectively carried out under three different wind power output ratios below, and three evaluation indicators under different wind power ratios are calculated according to the evaluation results as Figure 4 shown. From Figure 4 the A CC indicator, it can be seen that the proposed method has good prediction accuracy for different wind power ratios. The smaller the number of steps, the higher the prediction accuracy of the error category. The A CC of the same prediction step under the three wind power ratios

[0157] is not much different, but the larger the wind power ratio, the larger the fluctuation range of the wind power error, making the prediction difficulty of its error prediction trend relatively increase. Figure 4 compares the 1-16 step interval coverage rate I PICP obtained by the proposed model under different wind power ratio scenarios with the optimal I PICP in some existing literatures to reflect the reliability of the model proposed by the present invention. PICP It is found that the interval coverage rate I PICP obtained by the method proposed by the present invention is basically higher than that of the existing literature under different numbers of steps. The I PICP of the 16-step prediction under the three wind power ratios can basically reach about 0.9, and the I PICP of the 1-step prediction can even reach 1.0, indicating that the error classification method proposed by the present invention has a better recognition and prediction effect on the error category, making the actual wind power data basically included in the evaluation interval range. In addition, by comparing the I PICP under different numbers of steps under the same wind power ratio, it is found that there is also a basic trend that the smaller the prediction step, the larger the I

[0158] In addition, Figure 4 also compares the 1-16 step interval average bandwidth I PINAW obtained by the proposed model with the optimal I PINAWResults. It can be seen that, on the premise that the proportion of wind power output remains unchanged, the average bandwidth of the prediction interval has no obvious change trend in steps 1-16. Compared with the method of improving the prediction interval solely relying on the prediction model, the proposed method is less affected by the prediction steps. This is because the present invention classifies the errors by using the spare power that the system can provide. For the same type of prediction error, the upper and lower limits remain the same at different steps, and the fluctuation range remains unchanged. In addition, at the same prediction step, the larger the proportion of wind power output, the smaller its I PINAW is, but the increase in the proportion of wind power does not make A CC and I PICP decrease. This also reflects that in the future when the proportion of wind power is getting higher and higher, the three evaluation indicators of the proposed interval evaluation method for classifying errors by using the spare power of the system are all excellent, which is effective for improving the wind power prediction interval effect.

[0159] It can be seen that most of the existing literature studies the wind power interval prediction effect at a certain specific prediction step, especially the 1-step, 4-step and 16-step predictions. In the future, the proportion of wind power is getting higher and higher, and the scheduling strategy with a fixed scheduling cycle of the system is no longer applicable. However, dynamically adjusting the scheduling cycle depends on the prediction errors at different prediction steps of wind power. Therefore, taking the 1-16 step predictions as an example, the present invention explores the influence of the prediction step on the wind power prediction error, which also helps to provide a basis for adjusting the system scheduling cycle.

[0160] Analysis of the types of wind power prediction errors:

[0161] According to the wind power prediction results, the classification of the 1-16 step prediction errors of each sample point in the test set when the proportion of wind power is 16% is shown in Figure 5(a). According to the illustrated wind power prediction error categories, it can be found that when the wind power fluctuates gently, the wind power prediction errors basically belong to Class III. When the wind power surges or drops suddenly (such as 250-310 point, 620-660 point, etc.), the wind power prediction errors are larger and belong to Class II or Class IV. However, since the proportion of wind power in the scheduling system is small at this time and the conventional units can provide a large amount of spare power, there are no Class I or Class V prediction errors in this scenario.

[0162] In fact, the smaller the proportion of wind power, the stronger the system's compensation ability for its uncertainty. At this time, the system's available reserve power is used to classify the errors, and extreme types are less likely to occur. Therefore, compared with the prediction results when the wind power proportion is 16%, more of the prediction errors when the wind power proportion is 11% belong to Class III, and fewer belong to Class II or Class IV. No special plot analysis is presented here. On the contrary, when the proportion of wind power is larger, the system's compensation ability for its uncertainty weakens, and extreme situations are likely to occur. Therefore, the classification of the 1- to 16-step prediction errors of each sample point in the test set when the wind power proportion is 20% is plotted for analysis, as shown in Fig. 5(b). Fig. 5(b) has the same pattern as Fig. 5(a). However, compared with Fig. 5(a), more of the prediction error types in the sharp rise and fall periods of wind power in Fig. 5(b) (such as 90 - 150 point, 400 - 440 point, etc.) belong to Class II or Class IV. Even the 16-step prediction error types at the sample points in the peak period (105 - 110 point, 427 - 430 point) belong to Class I or Class V, indicating that the rapid fluctuation of wind power in a short time increases the prediction difficulty at this time, and the excessive prediction error causes the system to be unable to compensate.

[0163] Combining Fig. 5(a) and Fig. 5(b), it can be seen that overall, for the same sample point, the larger the prediction step or the greater the short-term fluctuation of wind power, the more likely the wind power prediction error type is to belong to Class II or Class IV, or even Class I or Class V, and the worse the system's compensation ability for the uncertainty of wind power. Therefore, it is necessary to adjust the system's scheduling cycle to improve the system's stability.

[0164] Analysis of scheduling results:

[0165] Using the evaluation results of the 1- to 16-step wind power prediction error intervals of the above sample points can guide the appropriate adjustment of the scheduling cycle, enabling the system to use its own reserve power to compensate for the prediction errors at any scheduling point and reducing the adverse impact of wind power uncertainty on the system. In scenarios with different wind power proportions, 96 sample points are selected respectively to simulate the intraday scheduling verification of a day, and the following two cases are set to compare and verify the impact of this method on the adjustment of the scheduling cycle:

[0166] CaseⅠ: Set the wind power fluctuation interval with a 90% confidence interval and a fixed 4-hour scheduling cycle;

[0167] CaseⅡ: Determine the wind power fluctuation interval according to the prediction error category and dynamically adjust the scheduling cycle;

[0168] The adjustment results of the scheduling cycle under different wind power ratios are shown in Figures 6(a), 6(b) and 6(c). As can be seen from Figures 6(a), 6(b) and 6(c), if the wind power prediction error types from a certain moment to the last moment are all type II or type IV in the 4 hours from the current moment to the future, or the wind power prediction error type at a certain moment is type I or type V, then adjust the scheduling cycle before that moment, start a new scheduling period at that moment, and shorten the prediction steps at that moment, so as to reduce the wind power prediction error, enabling the system to sufficiently utilize its own reserve power to compensate for the wind power prediction error and improving the system stability.

[0169] According to the adjustment results of the scheduling cycle, the scheduling results of Case I and Case II are analyzed respectively under different wind power ratios, and the system scheduling results are shown in Table 2. As can be seen from Table 2, regardless of the wind power ratio, after dynamically adjusting the scheduling cycle according to the error category, the system reduces both the wind curtailment and load shedding while reducing the total scheduling cost, improves the system's ability to cope with wind power uncertainty, and at the same time improves the economy of the system.

[0170] Table 2 Comparison of Scheduling Results of Case I and Case II under Different Wind Power Ratios

[0171]

[0172] Analysis of Wind Power Data Results of Domestic Wind Farms:

[0173] To further prove the effectiveness of the strategy of the present invention, the data of domestic wind farms are also used for verification. Table 3 gives three evaluation indicators after interval prediction of domestic wind power data, and the classification of the 1-16 step wind power prediction errors of each sample point in its test set is plotted as Figure 7 shown.

[0174] Table 3 Evaluation Indicators for Prediction Intervals of Domestic Wind Power Data

[0175] Prediction step size <![CDATA[A CC > <![CDATA[I PICP > <![CDATA[I PINAW > 1 step 0.967 0.991 0.4208 2 steps 0.923 0.979 0.4239 3 steps 0.866 0.940 0.4270 4 steps 0.849 0.920 0.4243 5 steps 0.813 0.885 0.4221 6 steps 0.817 0.898 0.4122 7 steps 0.778 0.846 0.3980 8 steps 0.745 0.836 0.3938 9 steps 0.764 0.831 0.4075 10 steps 0.739 0.831 0.4058 11 steps 0.706 0.816 0.4058 12 steps 0.712 0.835 0.4021 13 steps 0.694 0.787 0.3963 14 steps 0.676 0.788 0.3685 15 steps 0.667 0.784 0.3755 16 steps 0.664 0.784 0.3796

[0176] Compared with the data provided by Elia, the uncertainty of domestic wind farm data is greater and the volatility is stronger under similar ratios, making its error change trend more difficult to predict. Therefore, compared with Figure 4 at the same prediction step, the evaluation accuracy Acc of domestic wind farm data is relatively low, which is also related to the source of the original prediction error, that is, the original wind power prediction method. However, its interval coverage rate I PICP is still relatively high. According to the calculation, the ratio of this wind power data is about 10%, but its I PINAW is less than the I of this data of Elia data when the wind power ratio is 11% PINAWAs a result, since the proportions of the two groups of data participating in the system's historical scheduling are similar and the available standby power provided by the system does not differ much, the interval bandwidths obtained after classification are basically the same. However, the volatility of domestic wind power data is relatively large, resulting in a large range R for this dataset. Therefore, its I PINAW is smaller.

[0177] It can be seen from Figure 7 that the uncertainty of domestic wind farm data is large. Even though the proportion of its wind power is small, a large number of type II or type IV wind power prediction errors still appear in the sample points, and even type I or type V prediction errors occur. In addition, the frequent fluctuations of its wind power result in the 2-step and 3-step wind power prediction error types of sample points in some time periods (such as 170 - 310 point, 360 - 410 point, 745 - 785 point, etc.) already belonging to type II or type IV, and even the 14-step wind power prediction error type of individual sample points is type I or type V. It can be seen that even if the proportion of wind power is small, if the fluctuations of wind power are too frequent, its prediction error will be large. Even when the prediction step is small, the system's compensation ability relying only on its own standby power is still insufficient, further reflecting the necessity of the strategy of the present invention.

[0178] Therefore, according to the evaluation results of the 1 - 16-step wind power prediction error intervals of each sample point of domestic wind farms, 96 sample points per day are still selected for the comparative scheduling verification of Case I and Case II. The adjusted results of the scheduling cycle are as Figure 8 shown. It can be seen that even when the proportion of wind power is relatively low at this time, the strong volatility of wind power causes the adjustment of the system's scheduling cycle to be relatively frequent in order to ensure the stability of system scheduling.

[0179] According to Figure 8 the adjusted results of the scheduling cycle of domestic wind farms shown, the scheduling results of Case I and Case II are analyzed, and the system scheduling results are shown in Table 4. The results show that since there are periods with approximately zero output of wind power in this scenario, the load shedding after scheduling increases by nearly 39 MW. However, after dynamically adjusting the scheduling cycle using the method of the present invention, the curtailment of wind power in Case II is reduced by 285.652 MW. Overall, the wind power consumption capacity is still significantly improved, and the total system scheduling cost is also reduced, improving the economy of the system.

[0180] Table 4 Comparison of Scheduling Results of Case I and Case II under Domestic Data

[0181] Case Total system cost / $ Wind curtailment / MW Load shedding / MW CaseⅠ <![CDATA[2.395×10 6 > 1201.789 327.942 CaseⅡ <![CDATA[2.389×10 6 > 916.137 366.933

[0182] According to the wind power prediction error interval evaluation method based on system dispatching reserve classification proposed by the embodiments of the present invention, first, historical wind power data is utilized to participate in system dispatching, and the wind power prediction errors are classified according to the reserve power range obtained from the system dispatching. Then, taking the historical wind power data and historical prediction errors as the model inputs and the prediction error interval as the output, an error interval evaluation model is established to analyze the influence of the prediction step length and wind power waveform on the types of wind power prediction errors. Finally, based on the obtained types of prediction errors, the dispatching period of the system is dynamically adjusted to improve the reliability of system dispatching.

[0183] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or N embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0184] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of the present invention, the meaning of "N" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0185] Any process or method description in the flowchart or described in other ways herein can be understood as representing a module, segment, or part of code including one or N executable instructions for implementing a customized logical function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in a reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the art of the embodiments of the present invention.

Claims

1. A wind power forecast error interval evaluation method based on system dispatch reserve classification, characterized in that: The following steps are involved: Step 1, establish an available reserve model for the integrated energy system based on historical wind power; Step 2, establishing a wind power forecast error classification model based on system available reserve; Step 3: According to the available reserve model of the integrated energy system and the wind power prediction error classification model, a wind power prediction error interval evaluation model based on the system scheduling reserve classification is established, the wind power prediction error type is predicted using the wind power prediction error interval evaluation model, and the scheduling cycle of the integrated energy system is adjusted according to the predicted wind power prediction error type.

2. The method according to claim 1, characterized in that Step 1 specifically includes the following steps: Step 1.1: Establish the comprehensive energy system cost objective function C and solve it separately: minC=min(C CON +C CHP +C QF ) Where: C is the total system cost, C CON is the operating cost of thermal power units, C CHP is the operating cost of the combined heat and power unit, C QF Penalty costs for wind curtailment; Where: N1 is the number of thermal power units, N2 is the number of cogeneration units, t total is the total scheduling period, a k 、b k 、c k is the operating cost coefficient of thermal power unit k, α m , β m , γ m , δ m ,θ m , μ m is the operating cost coefficient of thermal power unit m, c1 is the price of thermal coal, ζ QF is the wind abandonment penalty coefficient, P CON,k,t is the power output of thermal power unit k at time t, P CHP,m,t is the power output of cogeneration unit m at time t, Q CHP,m,t is the heat output of CHP unit m at time t, P wind,t is the wind power output at time t, and They are the upper and lower limits of wind power output at time t respectively; Step 1.2: Constraints that must be satisfied when solving the cost objective function C: The system power balance constraints are divided into electrical power balance and thermal power balance, where the electrical power balance and thermal power balance constraints are as follows: Where: P LD,t is the total power of the electric load at time t, P n,t is the storage / discharge power of the storage device, P n,t >0 indicates power storage, P n,t <0 indicates discharge, Q LD,t is the total heat load power at time t; The unit operation constraints mainly include unit output constraints, unit ramp constraints and thermal-electric coupling constraints of cogeneration units. Among them, the output constraints of thermal power units and cogeneration units are as follows: Where: P CON is the power of the thermal power unit, and are the upper and lower limits of the power of thermal power units respectively; P CHP is the electrical power of the combined heat and power unit, and are the upper and lower limits of the combined heat and power unit power, Q CHP is the thermal power of the cogeneration unit, and are the upper and lower limits of the thermal power of the cogeneration unit, P wind To contribute to wind power, and They are the upper and lower limits of wind power output respectively; The ramp constraints for thermal power units and cogeneration units are as follows: Where: and are the landslide power and climbing power of the thermal power unit, and They are the landslide power and climbing power of the cogeneration unit respectively; The combined heat and power unit has thermal and electrical coupling constraints, which are expressed as follows: Where: C m is the heat-to-electricity ratio of the cogeneration unit m under back pressure conditions, c m is the heat-to-electricity ratio of the combined heat and power unit m under the maximum condensing condition, K m is a constant; The constraints of the power storage device are divided into two categories: one is the constraints on the storage and discharge instructions X, and the other is the constraints on the storage and discharge power and the capacity of the power storage device: Where: and are the storage instructions and discharge instructions of the power storage device at time t, and is the storage power and discharge power of the storage device, and is the maximum storage and discharge power and the minimum storage and discharge power that can be provided in a single dispatch period, E n,t , and is the storage capacity, maximum available capacity and minimum available capacity of the storage device at time t, η chu and η f is the charging efficiency and discharging efficiency of the power storage device; Step 1.3: Solve the available reserve model for the integrated energy system: The available reserve of the integrated energy system includes the total reserve power provided by the conventional units and the reserve power provided by the power storage device after the dispatch is completed. It is divided into the reserve power and the reserve power, which can be expressed as: Where: and are the total reserve power and total standby power that the system can provide at time t, respectively. and are the reserve power and standby power provided by the thermal power unit at time t, and The reserve power and standby power provided by the cogeneration unit at time t, and are the reserve power and standby power provided by the power storage device at time t, It is the maximum storage and discharge power that the power storage device can provide.

3. The method according to claim 2, characterized in that Step 2 specifically includes: considering the correlation between the system storable / dispensable reserve power curve and the positive and negative prediction errors of wind power, the wind power prediction errors are divided into five categories: The wind power prediction error is positive and greater than the maximum value of the system's available reserve power R a,fang,max When , the current wind power prediction error is set to Class I; When the wind power prediction error is positive and between the maximum and minimum values ​​of the system's available reserve power, the current wind power prediction error is set to Class II; The wind power prediction error is positive and less than the minimum reserve power R that the system can put a,fang,min , or the wind power forecast error is negative and its absolute value is less than the minimum reserve power R of the system a,chu,min When , the current wind power prediction error is set to Class III; When the wind power prediction error is negative and its absolute value is between the maximum and minimum values ​​of the system's reserve power, the current wind power prediction error is set to Class IV; The wind power forecast error is negative and its absolute value is greater than the maximum value of the system's reserve power R a,chu,max When , the current wind power prediction error is set to Class V.

4. The method according to claim 3, characterized in that Step 3 specifically includes the following steps: Step 3.1: Using LSTM neural network, establish a wind power forecast error interval evaluation model based on system dispatch reserve classification; The LSTM neural network consists of an input layer, a recursive hidden layer, and an output layer. The reading and modification of memory cells in the LSTM are achieved by controlling the input gate, forget gate, and output gate. The specific calculation formula is as follows: Where: W xc , W xi , W xf , W xo For connecting input signal x t The weight matrix, W hc , W hi , W hf , W ho is the output signal h of the connected hidden layer t The weight matrix, W ci , W cf , W co Output vector c for the connected neuron activation function t The diagonal matrix of the sum gate function, b i 、b c 、b f 、b o is the bias vector, σ is the activation function; Taking wind power data and wind power prediction error as input and future wind power prediction error type and interval as output, a wind power prediction error interval evaluation model is established by using the association between the input and output of the LSTM neural network. The expression is: h t+1 =f(x t ,x t-1 ,…,x t-n ) Where: h t+1 is the prediction error type, x i is the input data, which is the wind power data P i and wind power forecast error data e i Combination, x i =(P i , e i ), i = t, t-1, ..., tn; Step 3.2: Solve the evaluation index of the wind power forecast error interval evaluation model: The reliability of the wind power forecast error interval assessment model is verified by using three evaluation indicators: assessment error category accuracy, assessment interval coverage, and interval average bandwidth: (1) Evaluate the accuracy of error categories A CC : Where: N is the number of evaluation samples. When the original error category and the evaluation error category are the same, k i is 1, otherwise it is 0; A CC The value range is [0,1], which represents the accuracy of the evaluation model. CC The closer it is to 1, the higher the evaluation accuracy. CC The smaller it is, the lower the accuracy is and the worse the interval evaluation effect is; (2) Evaluation interval coverage I PICP : Where: If the evaluation target value falls within the evaluation interval, c i is 1, otherwise it is 0; the greater the frequency of the actual wind power value falling within the evaluation interval, that is, the closer the interval coverage rate is to 1, the better the interval prediction effect is; (3) Interval average bandwidth I PINAW : The interval width δ of the evaluation interval is the difference between the upper bound U and the lower bound L, that is: δ=UL The average bandwidth of the interval is: Where: R is the difference between the maximum and minimum values ​​of the test sample, I PINAW The smaller it is, the narrower the prediction interval is; Step 3.3: According to the obtained prediction error interval type, determine whether the integrated energy system needs to adjust the scheduling cycle at the current moment.

5. The method according to claim 4, characterized in that The method for adjusting the dispatch cycle of the integrated energy system according to the predicted wind power forecast error type is as follows: If the wind power prediction errors at the corresponding prediction steps at each time point in the next 4 hours are all Class III, the dispatch plan for the next 4 hours is directly formulated at the current time; If only some discontinuous time points have wind power forecast errors of Class II or Class IV under the corresponding forecast steps within 4 hours from the current moment, the dispatch plan for the next 4 hours is still formulated at the current moment; If the wind power prediction errors at the corresponding prediction steps at each time point in the preset time period in the future are all Class III, and the wind power prediction errors at the corresponding prediction steps at each time point in the future 2h-4h are all Class II or Class IV, adjust the dispatch cycle of the integrated energy system at the current moment and formulate a dispatch plan for the next 2h; If the prediction error at a certain time point corresponding to the prediction number of steps within 4 hours after the current time point is Class I or Class V, a new scheduling instruction is issued before the current time point and a new scheduling cycle is entered.

Images