Electric energy router energy coordination control method

Abstract

The invention relates to the technical field of intelligent power grids, in particular to an electric energy router energy coordination control method, which comprises the following steps of performing real-time prediction on the power condition of each unit of an electric energy router by utilizing a two-way long-short-term memory network prediction model based on a sparrow algorithm based on historical data; according to predicted values of basic loads and flexible loads (including translational loads, transferable loads and subtractable loads), a market price adjustment mechanism is considered, and a flexible load scheduling method based on time-of-use electricity price is provided with the purpose of minimum operation cost. The invention provides an electric energy router energy coordination control method based on an improved differential evolution algorithm of a random adaptive operator. According to the invention, under the framework of a market price adjustment mechanism, the economical efficiency of energy coordination control of the electric energy router is effectively improved, and meanwhile, the regulation and control capability of the market price to the power load is remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of smart grids, and in particular to an energy coordination control method for an electric energy router. Background Art

[0002] As photovoltaic power generation, wind power, hydropower and other new energy sources account for an increasing proportion in the power industry, the intermittent and volatile nature of their power generation leads to a temporal and spatial imbalance in the supply and demand of electricity, and wind and solar power abandonment occurs from time to time. Establishing and improving the price mechanism of the power auxiliary service market and giving full play to the price adjustment mechanism of the power market to flexibly dispatch flexible loads are important means to alleviate the contradiction between power supply and demand, reduce the phenomenon of wind and solar power abandonment, and enhance the local consumption of new energy. Therefore, flexible dispatching of flexible loads through time-of-use electricity prices can reduce the phenomenon of wind and solar power abandonment and promote the improvement of the utilization rate of new energy.

[0003] On the other hand, as a core power management device for smart grids, the power router has multiple interfaces such as basic load, flexible load, energy storage, photovoltaic, wind power, and large power grid, realizing functions such as new energy generation, grid connection, system monitoring, and distribution protection. It connects the microgrid system to the distribution network in a local access manner. Under the premise of ensuring the quality of power, the power router can flexibly and actively manage the dynamic power within the regional power grid, and cooperate with other power routers in the distribution network to realize the entire intelligent power management. With the rise of energy Internet technology, the power router plays a key role in optimizing the allocation of power resources, improving the transmission efficiency of energy networks, promoting the consumption of new energy, and reducing the carbon emissions of the system.

[0004] At present, the power management and control of power routers are all determined by the energy coordination management layer, which gives the given values ​​of each power unit control system. The impact of the market price regulation mechanism on the coordinated control of flexible loads is rarely considered, resulting in insufficient regulation ability of the market price regulation mechanism on the power system and poor dynamic performance and economy of power distribution.

[0005] A method for energy coordination control of a multi-port power router authorized by a comparative document (CN112018825B). The method includes: providing a method for energy coordination control of a multi-port power router, wherein the method obtains a first correction value of the output power reference value of the energy storage battery power unit and a first correction value of the power consumption reference value of the power distribution network interface power unit through an energy scheduling optimization algorithm; obtains a second correction value of the output power reference value of the energy storage battery power unit, a second correction value of the power consumption reference value of the power distribution network interface power unit and dynamic load demand through an energy balance coordination control algorithm; and corrects the output power reference value of each power unit's own controller based on three variables: the power deviation between each energy supply power unit and the load unit in the power router, the system DC bus voltage deviation and the dynamic change of the load, thereby improving the stability and dynamic performance of the system.

[0006] Although the above patent realizes a multi-port energy coordination control of the power router, only the dynamic changes of the load demand are considered in the energy coordination control process, and the influence of the market price adjustment mechanism on the flexible load coordination control is not considered. In the energy coordination control process of the power router of the present invention, the power conditions of each unit of the power router are predicted in real time based on historical data using a bidirectional long short-term memory network prediction model (BiLSTM) based on the sparrow algorithm; according to the predicted values ​​of the basic load and the flexible load (including the shiftable load and the transferable load), the market price adjustment mechanism is considered, and a flexible load scheduling method based on time-of-use electricity price is proposed for the purpose of minimizing the operating cost; based on the improved differential evolution algorithm of the random adaptive operator, an energy coordination control method of the power router is proposed, which effectively improves the economy of the energy coordination control of the power router under the framework of the market price adjustment mechanism, which is qualitatively different from the prior art, and an energy coordination control method of the power router is needed to solve this problem. Summary of the invention

[0007] The purpose of the present invention is to provide an energy coordination control method for an electric energy router to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: a method for energy coordination control of an electric energy router, comprising the following steps: S1. Considering that the power router has the function of managing the dynamic power within the regional power grid, based on the power router system structure, the power history data of each unit of the power router including wind power, photovoltaic power generation and other new energy power generation units are obtained as a sample set, the sample data is normalized, the normalization interval is [0,1], and the sample data set is dynamically randomly divided into a training set and a test set; S2. Based on the training data set and the test data set divided in step S1, a bidirectional long short-term memory (BiLSTM) neural network prediction model with dynamic hyperparameter settings is established based on the sparrow algorithm to perform real-time prediction of the power of each unit of the power router; S3. The load forecast value obtained in step S2 is analyzed based on the flexibility of time-of-use electricity prices and flexible loads without affecting the load demand. A flexible load output timing model based on time-of-use electricity prices is established for the purpose of minimizing operating costs. A flexible load scheduling method based on time-of-use electricity prices is proposed for the purpose of minimizing operating costs; S4. Based on the real-time power prediction value of each unit of the power router obtained in step S2, under the premise of source-load coordinated operation, based on the flexible load output timing of the time-of-use electricity price in step S3, considering the energy exchange between the power routers, a power router transaction cost model based on the flexible load of the time-of-use electricity price is established; S5. Based on the improved differential evolution algorithm with random adaptive operator, the power distribution of the power router and the distribution network is given, and the energy coordination control method of the power router is proposed.

[0009] As a preferred technical solution of the present invention, step S1 includes the following sub-steps: S1.1 According to the following formula, the input sample data set Normalization is performed, and the normalization interval is [0,1]: Where: is the normalized data, is the sample data, is the minimum value in the sample data, is the maximum value in the sample data; S1.2 The sample data set is calculated according to the following formula Perform dynamic random ratio division into training set and test set; The training set is: The test set is: Where: is the cardinality of set P, that is, the number of elements in set P, Generate a random number greater than 0 and less than 1, function Truncate the decimal part to the integer.

[0010] As a preferred technical solution of the present invention, step S2 includes the following sub-steps: S2.1 uses the Sparrow algorithm to optimize the learning rate, number of hidden layers, and maximum training rounds of the BiLSTM neural network prediction model; S2.2 sets the hyperparameters (learning rate, number of hidden layer 1, number of hidden layer 2 and maximum training rounds) of the BiLSTM neural network prediction model based on the optimal values ​​to improve the prediction accuracy and predict the real-time output of each unit of the power router.

[0011] As a preferred technical solution of the present invention, step S3 includes the following sub-steps: S3.1 One-day time-of-use electricity price model: Where: is the time-of-use electricity price vector for 24 hours a day; for The electricity price corresponding to the time period; S3.2 Shiftable load cost optimization algorithm: The time state sequence of the 24-hour shiftable load is recorded as: Where: is the translatable load time state vector for 24 hours a day; It is a 0-1 binary state variable. A value of 1 indicates that there is load shift at that moment, and vice versa, it indicates that there is no load shift at that moment. When the time interval of the translatable load is , then the power distribution that can be translated to negative is: Where: is the start time of the translatable load; is the power of the load that can be translated; is the duration of the translatable load, then the end time of the translatable load is , the load is , the time state sequence of 24 time periods , only when hour , other cases, ; The operating cost of a translatable load is: Where: the symbol "T" represents the transpose of a vector; If the time period that can be shifted within the scheduling cycle is ,but ,make ,but , the time interval of the shiftable load that can be shifted is , then the operating cost after load scheduling can be translated into in, , To make the function In the interval When the variable reaches its minimum value The value of is the time state vector after translation.

[0012] The translation scheduling decision is: (1) If , without performing translational dispatching behavior, the power distribution of the translational load does not change; (2) If , perform translational scheduling, and translate the power distribution of the load: ; (3) The operating cost of the movable load is: S3.3 Transferable load cost optimization algorithm: Because the transferable load can ensure that the total load is met during the entire cycle, the energy consumption in each period can be flexibly adjusted. The transferable load power distribution vector is: Where: , is a 0-1 binary state variable, It means that there is load transfer at this moment, and the transfer power is , otherwise, there is no load transfer.

[0013] Cost of transferable loads for: In order to prevent the equipment from frequently starting and stopping in different time periods, the start time after a single transferable load transfer is set to , and the end time is ,but The constraints should be met: Where: It is the minimum continuous operating time of the transferable load.

[0014] like , then the transfer power satisfies the constraint condition: Where: and They are i The upper and lower limits of the transfer capacity at all times.

[0015] According to the allowed power consumption time of the transferable load, the time period with lower electricity price can be selected in the time-of-use electricity price sequence for load transfer, and the transferable load power consumption optimization algorithm can be used to minimize the electricity cost while meeting the electricity demand. That is, a transferable load scheduling method based on time-of-use electricity price is proposed.

[0016] S3.4 Reducible load model: The reducible load has a certain tolerance for power reduction, time reduction or interruption of operation, and can be reduced accordingly according to scheduling needs.

[0017] Operating costs of load reduction: Where: The electricity price for load reduction compensation is, To reduce the time series, is a 0-1 binary state variable, and the load reduction distribution power is ; Satisfy the constraints: Where: and They are the upward and downward flexibility of the load that can be reduced.

[0018] The duration and number of reductions should meet the following constraints Where: and is the upper and lower limits of the cumulative reduction time. It is the maximum number of load shedding allowed. Generally speaking, if the distribution network can provide sufficient power, load shedding will not be easily performed.

[0019] As a preferred technical solution of the present invention, step S4 includes the following sub-steps: S4.1 Transaction cost model of power router based on time-of-use electricity price flexible load: Under the premise of giving priority to maximizing the consumption of renewable energy, the daily transaction cost of the router can be expressed as: Where: is the time series of the electric power purchased by the router from the distribution network, is the purchased electricity price series, is the time series of electric power sold by the router to the distribution network, is the electricity price series sold, (or ) is a 0-1 binary state quantity, (or ) indicates the purchase (or sale) of electricity from the distribution network. The power purchased is (or power sold is ), .

[0020] S4.2 In order to ensure the stable operation of the router, the power balance constraint based on the flexible load of time-of-use electricity price needs to be met. Where: Load power ; From sub-step S3.2, we can get: From sub-step S3.3, we can get: For the set {hydropower( hydro ),nuclear power( unclear )、Photothermal( csp ), pumped storage ( xx )、Wind power( wt )、Photovoltaic power generation( pv )}, that is, the total number of types of power generation modules, It is a 0-1 binary state quantity, and a value of 1 indicates the i The time period participates in power generation, otherwise it means that the time period does not participate in power generation; symbol It is a conjunction operator of logical operations, which ensures that the power router will not purchase and sell electricity at the same time. In addition, the sales and purchase amounts are both subject to the optimized shiftable load power, transferable load power and energy storage system output constraints.

[0021] S4.3 Energy storage system model based on price regulation In order to give priority to the consumption of renewable energy, strengthen the impact of market price mechanism on the charging and discharging strategy of energy storage system, and regulate the charging and discharging income and cost of energy storage system based on price: Where: is the time series of the energy storage system, is a 0-1 binary state variable, and the distributed power of the energy storage system is , Indicates charging, the energy storage system is equivalent to load, Indicates discharge, and the energy storage system is equivalent to a power source.

[0022] make , then the time series of the source-charge power difference is expressed as: .

[0023] when When , it means that the power supply exceeds the demand. At this time, the energy storage system can participate in charging, and the energy storage system itself is equivalent to the load. If the energy storage system cannot participate in charging, the power router can sell electricity to the distribution network during this period, so that the power router meets the power balance constraint.

[0024] when When , it means that the power demand is greater than the supply. At this time, the energy storage system can participate in the discharge. The energy storage system itself is equivalent to the power source. If the energy storage system cannot participate in the discharge behavior, the power router can purchase electricity from the distribution network during this period, so that the power router meets the power balance constraint.

[0025] As a preferred technical solution of the present invention, the improved differential evolution algorithm based on the random adaptive operator in step S5 includes the following sub-steps: S5.1 Initialize the population The population is defined as: , where D is the dimension of the solution space and the vector Indicates The "chromosomes" of the population. Indicates Daidi Chromosome No. Genes: In the formula, and Respectively represent chromosome The upper and lower bounds of the gene value range, . Represents the size of the population Is the interval The random number in .

[0026] S5.2 Mutation Operation The differential evolution algorithm uses the differential strategy to achieve individual mutation, that is, randomly select two different individuals in the population, scale their vector difference, and synthesize the vector with the individual to be mutated to generate an intermediate individual. : In the formula, is the best individual in the current population, is a random selection of individuals in the population, and . Random adaptive scaling factor It is a scaling factor whose value fluctuates and decreases as the number of iterations increases. The formula is as follows: In the formula, , represents the scaling factor The initial value of . represents the maximum number of iterations, Indicates the current number of iterations, function Represents the mean and standard deviation Generate random numbers from a normal distribution, generally , The value is relatively small to describe small random fluctuations and prevent the differential evolution algorithm from falling into a local optimal solution.

[0027] S5.3: Crossover Operations For Generation population and its variant intermediates Perform crossover operations between individuals: In the formula, For interval A random positive integer in is the crossover probability.

[0028] S5.4: Boundary constraints In the process of evolution, in order to ensure the validity of the solution, it is necessary to determine whether each "gene" in the "chromosome" satisfies the constrained boundary conditions. If the range of population individuals exceeds the given upper and lower limits, the boundary absorption method is used for boundary constraints: S5.5: Select an operation According to the greedy criterion, individuals entering the next generation population are selected: In the formula, For vector The fitness function is defined as follows: In the formula, is the daily transaction cost of the router. The specific expression is shown in S4.1 Power router transaction cost model. It is the power distribution of power routers for power trading and load reduction.

[0029] S5.6: Based on the optimization results, an internal optimization scheduling strategy for the power router is proposed to provide reference values ​​for power trading between the power router and the distribution network and the power distribution of the load that can be reduced.

[0030] As a preferred technical solution of the present invention, in step S2.2, the BiLSTM neural network prediction model based on the sparrow algorithm can be divided into the following sub-steps: S6.1 Sparrow Optimization Algorithm: The sparrow population is divided into discoverers, followers, and scouts. The discoverers are responsible for exploring new food sources and guiding the search direction of the group; followers search for food near the discoverers and conduct local searches; scouts observe whether there is any danger inside the sparrow group and remind the entire sparrow group to be safe.

[0031] Sparrow Collection Matrix as follows: Where: , It is the size of a sparrow. .

[0032] The fitness value of all sparrows is Where: , Each value in represents the fitness value of an individual, is the dimension of the variable.

[0033] It indicates a sparrow with a better fitness value, and this sparrow is the first to obtain food and, as a discoverer, leads the entire population toward the food source.

[0034] set up is a small positive number approaching zero, then the basis for judging whether the sparrow has reached the food source is: The position update formula of each generation of discoverers is as follows: Where: is the pre-set maximum number of iterations, for A uniform random number in is a standard normally distributed random number.

[0035] The position of each follower generation is updated as follows: Where: It is the sparrow with the worst fitness in the current population. ,in A row matrix where all elements are randomly set to 1 or -1.

[0036] Scouts need to keep an eye on the current environment so that they can immediately fly to a safe environment when they find danger. The position update is as follows: Where: is the sparrow with the best fitness in the current population, is a random number that conforms to the normal distribution. for A random number between to a smaller number to prevent the denominator from being 0. is the worst fitness value of the current population. is the global optimal fitness value.

[0037] S6.2 BiLSTM neural network prediction model: The single-layer BiLSTM in BiLSTM is composed of two continuous time step LSTMs, namely the forward continuous time step LSTM and the reverse continuous time step LSTM. Its core structure is a single-step LSTM network model, which includes three sigmoid activation functions. Keep the output range between 0 and 1. Each sigmoid activation function belongs to a flexible gate function. The gate function is divided into three different gate functions: forget gate , input gate , output gate ; a hyperbolic tangent function ; Input data timing , output data timing The three gate functions decide whether to forget or retain some or all of the inherited data through the current input and the output of the previous time step. The weight values ​​are learned through the back propagation algorithm. and deviation Narrow the gap between the predicted output and the actual training sample. Through this forget-retain-output iterative process, the current information can be retained, and the output is concatenated after forward and backward processing of the input sequence, and the final result is output.

[0038] The functional relationship of the single time step LSTM network prediction model is as follows: in , , , , , , and is the weight matrix of each LSTM network module. , , and is the bias in the LSTM network. represents the sigmoid activation function, Represents the hyperbolic tangent function.

[0039] S6.3 Optimal output vector according to the sparrow algorithm , set the hyperparameters of the BiLSTM neural network prediction model, such as the learning rate, number of hidden layers, and maximum training rounds, and predict the real-time output of each unit of the power router based on historical data.

[0040] As a preferred technical solution of the present invention, in step S3.3, the transferable load scheduling method based on time-of-use electricity price can be divided into the following sub-steps: S7.1 The cost objective function of the transferable load based on time-of-use electricity price is: S7.2 The constraints are: (1) The transfer power satisfies the constraints: in, and Respectively i The upper and lower limits of the transfer capacity at all times.

[0041] (2) The minimum continuous operation time constraint of the transferable load is: S.7.3 According to the improved differential evolution algorithm based on random adaptive operators as described in claim 5, electricity consumption is regulated based on time-of-use electricity prices. Under the conditions of the allowable electricity consumption time of the transferable load, the load output corresponding to the electricity consumption time period of the transferable load can be obtained, so as to minimize the electricity cost while meeting the electricity demand.

[0042] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention is an energy coordination control method for an electric energy router. The present invention is configured to dynamically divide a training set and a test set using a dynamic random ratio to improve prediction accuracy.

[0043] (2) The present invention is an energy coordination control method for an electric energy router. The present invention adopts a bidirectional long-short time memory neural network prediction model based on dynamic hyperparameter setting to perform high-precision real-time prediction of the power of each unit of the electric energy router.

[0044] (3) The present invention is a method for coordinated energy control of an electric energy router. The present invention proposes a flexible load dispatching method based on time-of-use electricity prices; the market price mechanism is included in the coordinated energy control to improve the economy of the control.

[0045] (4) The present invention is a method for coordinated energy control of an electric energy router. The present invention is based on an improved differential evolution algorithm of a random adaptive operator, incorporates a market price mechanism into the charging and discharging behavior of an energy storage system, and proposes a method for coordinated energy control of an electric energy router. The method has the advantages of promoting the consumption of new energy and reducing the carbon emissions of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 It is a flow chart of the method for regulating electric energy of an electric energy router based on price regulation of flexible loads according to the present invention; Figure 2 A BiLSTM neural network structure diagram of the power router power control method based on price-controlled flexible loads of the present invention; Figure 3 The present invention is a flow chart of the BiLSYM dynamic hyper-parameter setting based on the sparrow algorithm; Figure 4 It is a structural diagram of the power router of the present invention; Figure 5 It is a predicted waveform diagram of the power generation unit of the present invention; Figure 6 is a prediction error waveform diagram of the power generation unit of the present invention; Figure 7 This is a basic load prediction waveform diagram of the electric energy router of the present invention; Figure 8 This is a basic load prediction error waveform diagram of the electric energy router of the present invention; Fig. 9 This is the waveform diagram of the flexible load optimization result of the power router of the present invention on January 6; Fig.10 This is the waveform diagram of the flexible load optimization result of the power router of the present invention on January 8; Fig.11 The source-load waveform diagram of the present invention on January 6 with or without the energy storage system participating in the regulation; Fig.12 This is a source-load waveform diagram of the present invention with or without the energy storage system participating in regulation on January 8. DETAILED DESCRIPTION

[0048] Below, the invention is further described in conjunction with the accompanying drawings and specific embodiments: See also Figure 1-12, according to an energy coordination control method of an electric energy router according to an embodiment of the present invention, such as Figure 1 As shown, the method comprises the following steps: Step S1. Select the power router to be regulated, obtain the historical power data of each unit of the power router, and process the historical data, including the following sub-steps: S1.1 Select the power router to be regulated, such as Figure 3 In one embodiment of the present invention, time series historical data samples for 10 days and 240 hours from January 1 to 10, 2020 are collected and normalized; S1.2 divide the data processed in step S1.1 into a training sample set based on the 5-day 120-hour time series historical data set from January 1 to 5, 2020; and a test sample set based on the 5-day 120-hour time series historical data set from January 6 to 10, 2020; S1.3 summarizes the sample data obtained in all the above steps to form a model prediction sample data set.

[0049] Step S2. According to the model prediction sample data set obtained in step S1, a bidirectional long short-term memory (BiLSTM) neural network prediction model with dynamic hyperparameter settings is established based on the Sparrow algorithm to perform real-time prediction of the power of each unit of the power router; In step S2, if Figure 2 As shown, a dynamic hyperparameter setting flow chart of the BiLSTM neural network prediction model based on the sparrow optimization algorithm. Based on the hyperparameter setting flow chart, step S2 includes the following sub-steps: S2.1 provides training data for BiLSTM network training through the divided training sample set, and uses the mean square error as the fitness function to evaluate the training effect of BiLSTM network. The fitness function formula is as follows: Where: x is the true value; is the predicted value; is the expected function; S2.2 uses the sparrow optimization algorithm (SSA) to optimize the hyperparameters of the BiLSTM neural network prediction model. The number of iterations is set to 30, the population number n of the sparrow optimization algorithm is set to 5, and the dimension d is set to 4, which represent the number of nodes in hidden layer 1, the number of nodes in hidden layer 2, the maximum training rounds and learning rate, and the upper limit of model parameter search. , the search limit is ,The use of the Sparrow algorithm improves the prediction accuracy of the BiLSTM neural network prediction model; S2.3 The BiLSTM neural network prediction model based on the sparrow optimization algorithm predicts the output of each unit of the power router to be regulated. The prediction result waveform is shown in Figure 4-7 .

[0050] Step S3. According to the predicted values ​​of basic load, movable load and transferable load obtained in step S2, the forecast data on January 7, 2020 is selected, based on 3 time-of-use electricity prices: 0:00-8:00 valley section 0.37KW / h, 8:00-15:00 flat section 0.62KW / h, 15:00-23:00 peak section 0.93KW / h, 23:00-24:00 valley section 0.37KW / h, and the 24-hour power distribution of movable load and transferable load based on price-regulated flexible load is obtained.

[0051] In step S3, the 24-hour power distribution of the flexible load is regulated based on the price, including the following sub-steps: S3.1 Based on the predicted source-load power difference of the power router, the shiftable load power distribution is determined as: and shiftable time periods , based on the translation scheduling decision, the power distribution of the shiftable load after translation scheduling is obtained: S3.2 updates the source-load power difference of the power router and determines the transferable load power distribution as: According to the allowed power consumption time of the transferable load, the transferable load scheduling method based on the time-of-use electricity price selects the time period with lower electricity price in the time-of-use electricity price sequence for load transfer and updates the transferable load power distribution. .

[0052] Step S4. The hydropower generation, solar thermal generation, wind power, photovoltaic generation and basic load output time series forecast data of the power router in S2, as well as the 24-hour power time series data of the shiftable load and transferable load in step S3, are used to establish a power router operation cost model based on time-of-use electricity price flexible load, with the premise of giving priority to maximizing the consumption of renewable energy. The establishment of a single-day operation cost model of the power router includes the following sub-steps: The daily transaction cost function of S4.1 power router can be expressed as: Where: is the time series of the electric power purchased by the router from the distribution network, is the purchased electricity price series, is the time series of electric power sold by the router to the distribution network, is the electricity price series sold, (or ) is a 0-1 binary state quantity, (or ) indicates the purchase (or sale) of electricity from the distribution network. The power purchased is (or power sold is ), S4.2 Power balance constraints based on flexible loads based on time-of-use electricity prices Where: Load power ; From sub-step S3.1, we can get: From sub-step S3.2, we can get: All power sources participate in power generation; .

[0053] Step S5. Using an improved differential evolution algorithm based on a random adaptive operator to perform power coordination control on the power router includes the following sub-steps: S5.1 uses the predicted value obtained in step S3 as the basic data of the differential evolution algorithm and defines the population of the differential evolution algorithm, where , , the dimension of the solution space D =72, Indicates the real-time power of electricity purchase and sale. Indicates the real-time power of energy storage charging and discharging. Indicates the real-time power that can be used to reduce load.

[0054] S5.2 Introducing a random adaptive scaling factor in the mutation operation of the differential evolution algorithm Perform mutation operation, the formula is as follows: ; S5.3 performing a crossover operation, a boundary constraint, and a selection operation according to a differential evolution algorithm; S5.4 Repeat the steps S5.2 to S5.3, and finally obtain the optimal real-time power reference value of the power router for purchasing and selling electricity, storing energy, and reducing load.

[0055] Embodiment 1: This example selects an electric energy router as the research object, and the present invention provides an energy coordination control method for an electric energy router, and the steps are as follows: 1. First, process the historical data of the output of each unit of the power router to be regulated from January 1 to 10, 2020, and divide the training set and test set for the output prediction of each unit of the power router.

[0056] 2. The sparrow optimization algorithm (SSA) is used to optimize the hyperparameters of the BiLSTM neural network prediction model. The number of iterations is set to 30, the population number n of the sparrow optimization algorithm is set to 5, and the dimension d is set to 4, which represent the number of nodes in hidden layer 1, the number of nodes in hidden layer 2, the maximum training rounds and learning rate, and the upper limit of model parameter search. , the search limit is , obtain the BiLSTM neural network prediction model based on the Sparrow algorithm.

[0057] 3. According to the predicted values ​​of basic load, shiftable load and transferable load obtained in step S2, the forecast data of January 6 and January 8, 2020 are selected, and based on the three time-of-use electricity prices: 0.37KW / h in the valley section from 0:00 to 8:00, 0.62KW / h in the flat section from 8:00 to 15:00, 0.93KW / h in the peak section from 15:00 to 23:00, and 0.37KW / h in the valley section from 23:00 to 24:00, the 24-hour power distribution of the shiftable load and transferable load based on the price-regulated flexible load is obtained. Among them, on January 6, the power distribution of the shiftable load before optimization is The electricity cost is 550.74 yuan, and the power distribution before the transferable load optimization is The electricity cost is 515.14 yuan, and the power distribution after load optimization can be translated by the flexible load optimization algorithm proposed in this patent application: , the electricity cost is 400.10 yuan, and the power distribution before the transferable load optimization is: , the electricity cost is 289.34 yuan. On January 8, the power distribution before the load optimization is: , the electricity cost is 536.07 yuan, and the power distribution before the transferable load optimization is: The electricity cost is 496.76 yuan. By using the flexible load optimization algorithm proposed in this patent application, the power distribution after load optimization can be translated into The electricity cost is 388.23 yuan; the power distribution after the transferable load is optimized is: The cost is 276.76 yuan.

[0058] 4. The time series forecast data of hydropower generation, solar thermal power generation, wind power, photovoltaic power generation and basic load output of the power router in S2, as well as the 24-hour power time series data of the shiftable load and transferable load in step S3, are used. Under the premise of giving priority to maximizing the absorption of renewable energy, the energy storage system participates in the daily work of the power router, and the power exchange between the power router and the distribution network is considered to establish a transaction cost model of the power router based on time-of-use electricity price flexible load.

[0059] 5. An improved differential evolution algorithm based on random adaptive operators is used to coordinate the power of the power router. The energy storage system is selected to have a capacity of 3000kWh, a maximum output power of 600kW, and a maximum load power per unit time of 2000kW. Based on the above constraints, a method for energy coordination control of the power router is proposed. The power distribution of the power router and the distribution network transaction on January 8 is given as follows: During the period of 10:00~13:00 when the power distribution is 0, the power generation of the power router is greater than the load power consumption, and the energy storage system stores 304KWh of electricity as a power source during the period of 8:00~9:00, and acts as a load to store electricity during the period of 10:00~13:00 to absorb the power generation of new energy. In other time periods, the energy storage system shaves the peak of the daily load based on the time-of-use electricity price, and has a working power distribution of .

[0060] This example selects the electric routers (see Figure 3 ) of hydropower, solar thermal power, wind power, photovoltaic power and basic load output, collect the BiLSTM neural network prediction model based on the sparrow algorithm, and make real-time predictions on the power of each unit of the power router. For the specific process, see Figure 2 .pass Figure 4 , 5 The prediction results show that when the output time series data of each unit of the electric energy router from January 6 to 10 are selected as the test set, compared with the LSTM neural network prediction model based on the sparrow algorithm, the BiLSTM neural network prediction model based on the sparrow algorithm is used to perform time series prediction on the output of each unit of the electric energy router. The standard error of the prediction of solar thermal power output is reduced from 19.74 to 5.40, the standard error of the prediction of hydropower output is reduced from 0.98 to 0.62, the standard error of the prediction of wind power output is reduced from 76.68 to 43.91, and the standard error of the prediction of photovoltaic power output is reduced from 1.52 to 1.39. This shows that the BiLSTM prediction model based on the sparrow algorithm proposed in the present invention has higher prediction accuracy.

[0061] pass Figure 6From the results of flexible load optimization, it can be seen that after selecting three time-of-use electricity prices, the flexible load optimization algorithm based on price regulation is used to optimize the shiftable load and transferable load on January 6 and January 8 respectively. Among them, the shiftable load cost on January 6 was reduced by 150.64 yuan before and after optimization, and the shiftable load cost was reduced by 225.80 yuan before and after optimization. The shiftable load cost on January 6 was reduced by 147.84 yuan before and after optimization, and the shiftable load cost was reduced by 220.00 yuan before and after optimization. This shows that the optimization algorithm based on price-regulated flexible load proposed in the present invention can reduce the operating cost of flexible load and increase the regulation ability of the electricity price market mechanism.

[0062] pass Figure 7 From the results of the energy storage system participating in the coordinated control of electric energy, it can be seen that the daily transaction cost model of the electric energy router is combined with the output model of the energy storage system with price regulation, and the improved differential evolution algorithm based on random adaptive operators is used for optimization. This can reduce the power consumption of the electric energy router during peak hours. This shows that the energy coordinated control method of the electric energy router proposed in the present invention can strengthen the electricity price market mechanism and help promote the establishment of a sound power auxiliary service market.

[0063] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for energy coordination control of an electric energy router, characterized in that: The following steps are involved: S1. Considering that the power router has the function of managing the dynamic power within the regional power grid, based on the power router system structure, the power history data of each unit of the power router including wind power, photovoltaic power generation and other new energy power generation units are obtained as a sample set, the sample data is normalized, the normalization interval is [0,1], and the sample data set is dynamically randomly divided into a training set and a test set; S2. Based on the training data set and the test data set divided in step S1, a bidirectional long short-term memory (BiLSTM) neural network prediction model with dynamic hyperparameter settings is established based on the sparrow algorithm to perform real-time prediction of the power of each unit of the power router; S3. The load forecast value obtained in step S2 is analyzed based on the flexibility of time-of-use electricity prices and flexible loads without affecting the load demand. A flexible load output timing model based on time-of-use electricity prices is established for the purpose of minimizing operating costs. A flexible load scheduling method based on time-of-use electricity prices is proposed for the purpose of minimizing operating costs; S4. Based on the real-time power prediction value of each unit of the power router obtained in step S2, under the premise of source-load coordinated operation, based on the flexible load output timing of the time-of-use electricity price in step S3, considering the energy exchange between the power routers, a power router transaction cost model based on the flexible load of the time-of-use electricity price is established; S5. Based on the improved differential evolution algorithm with random adaptive operator, the power distribution of the power router and the distribution network is given, and the energy coordination control method of the power router is proposed.

2. The energy coordination control method of an electric energy router according to claim 1, characterized in that: The step S1 comprises the following sub-steps: S1.1 According to the following formula, the input sample data set Normalization is performed, and the normalization interval is [0,1]: Where: is the normalized data, is the sample data, is the minimum value in the sample data, is the maximum value in the sample data; S1.2 The sample data set is calculated according to the following formula Perform dynamic random ratio division into training set and test set; The training set is: The test set is: Where: is the cardinality of set P, that is, the number of elements in set P, Generate a random number greater than 0 and less than 1, function Truncate the decimal part to the integer.

3. The energy coordination control method of an electric energy router according to claim 1 is characterized in that: The step S2 comprises the following sub-steps: S2.1 uses the Sparrow algorithm to optimize the learning rate, number of hidden layers, and maximum training rounds of the BiLSTM neural network prediction model; S2.2 sets the hyperparameters (learning rate, number of hidden layer 1, number of hidden layer 2 and maximum training rounds) of the BiLSTM neural network prediction model based on the optimal values ​​to improve the prediction accuracy and predict the real-time output of each unit of the power router.

4. The energy coordination control method of an electric energy router according to claim 1, characterized in that: The step S3 comprises the following sub-steps: S3.1 One-day time-of-use electricity price model: Where: is the time-of-use electricity price vector for 24 hours a day; for The electricity price corresponding to the time period; S3.2 Optimization algorithm for shiftable load cost: The time state sequence of the 24-hour shiftable load is recorded as: Where: is the translatable load time state vector for 24 hours a day; It is a 0-1 binary state variable. A value of 1 indicates that there is load shift at that moment, and vice versa, it indicates that there is no load shift at that moment. When the time interval of the translatable load is , then the power distribution that can be translated to negative is: Where: is the start time of the translatable load; is the power of the load that can be translated; is the duration of the translatable load, then the end time of the translatable load is , the load is , the time state sequence of 24 time periods , only when hour , other cases, ; The operating cost of a translatable load is: Where: the symbol "T" represents the transpose of a vector; If the time period that can be shifted within the scheduling cycle is ,but ,make ,but , the time interval of the shiftable load that can be shifted is , then the operating cost after load scheduling can be translated into in, , To make the function In the interval When the variable reaches its minimum value The value of is the time state vector after translation. The translation scheduling decision is: (1) If , without performing translational dispatching behavior, the power distribution of the translational load does not change; (2) If , perform translational scheduling, and translate the power distribution of the load: ； (3) The operating cost of the movable load is: S3.3 Transferable load cost optimization algorithm: Because the transferable load can ensure that the total load is met during the entire cycle, the energy consumption in each period can be flexibly adjusted. The transferable load power distribution vector is: Where: , is a 0-1 binary state variable, It means that there is load transfer at this moment, and the transfer power is , otherwise, there is no load transfer. Cost of transferable loads for: In order to prevent the equipment from frequently starting and stopping in different time periods, the start time after a single transferable load transfer is set to , and the end time is ,but The constraints should be met: Where: It is the minimum continuous operating time of the transferable load. like , then the transfer power satisfies the constraint condition: Where: and They are i The upper and lower limits of the transfer capacity at all times. According to the allowed power consumption time of the transferable load, the time period with lower electricity price can be selected in the time-of-use electricity price sequence for load transfer, and the transferable load power consumption optimization algorithm can be used to minimize the electricity cost while meeting the electricity demand. That is, a transferable load scheduling method based on time-of-use electricity price is proposed. S3.4 Reducible load model: The reducible load has a certain tolerance for power reduction, time reduction or interruption of operation, and can be reduced accordingly according to scheduling needs. Operating costs of load reduction: Where: The electricity price for load reduction compensation is, To reduce the time series, is a 0-1 binary state variable, and the load reduction distribution power is ; Satisfy the constraints: Where: and They are the upward and downward flexibility of the load that can be reduced. The duration and number of reductions should meet the following constraints Where: and is the upper and lower limits of the cumulative reduction time. It is the maximum number of load shedding allowed. Generally speaking, if the distribution network can provide sufficient power, load shedding will not be easily performed.

5. The energy coordination control method of an electric energy router according to claim 1, characterized in that: The step S4 comprises the following sub-steps: S4.1 Transaction cost model of power router based on time-of-use electricity price flexible load: Under the premise of giving priority to maximizing the consumption of renewable energy, the daily transaction cost of the router can be expressed as: Where: is the time series of the electric power purchased by the router from the distribution network, is the purchased electricity price series, is the time series of electric power sold by the router to the distribution network, is the electricity price series sold, (or ) is a 0-1 binary state quantity, (or ) indicates the purchase (or sale) of electricity from the distribution network. The power purchased is (or power sold is ), . S4.2 In order to ensure the stable operation of the router, the power balance constraint based on the flexible load of time-of-use electricity price needs to be met. Where: Load power ; From sub-step S3.2, we can get: From sub-step S3.3, we can get: For the set {hydropower( hydro ),nuclear power( unclear )、Photothermal( csp ), pumped storage ( xx )、Wind power( wt )、Photovoltaic power generation( pv )}, that is, the total number of types of power generation modules, It is a 0-1 binary state quantity, and a value of 1 indicates the i The time period participates in power generation, otherwise it means that the time period does not participate in power generation; symbol It is a conjunction operator of logical operations, which ensures that the power router will not purchase and sell electricity at the same time. In addition, the sales and purchase amounts are both subject to the optimized shiftable load power, transferable load power and energy storage system output constraints. S4.3 Energy storage system model based on price regulation In order to give priority to the consumption of renewable energy, strengthen the impact of market price mechanism on the charging and discharging strategy of energy storage system, and regulate the charging and discharging income and cost of energy storage system based on price: Where: is the time series of the energy storage system, is a 0-1 binary state variable, and the distributed power of the energy storage system is , Indicates charging, the energy storage system is equivalent to load, Indicates discharge, and the energy storage system is equivalent to a power source. make , then the time series of the source-charge power difference is expressed as: . when When , it means that the power supply exceeds the demand. At this time, the energy storage system can participate in charging, and the energy storage system itself is equivalent to the load. If the energy storage system cannot participate in charging, the power router can sell electricity to the distribution network during this period, so that the power router meets the power balance constraint. when When , it means that the power demand is greater than the supply. At this time, the energy storage system can participate in the discharge. The energy storage system itself is equivalent to the power source. If the energy storage system cannot participate in the discharge behavior, the power router can purchase electricity from the distribution network during this period, so that the power router meets the power balance constraint.

6. The energy coordination control method of an electric energy router according to claim 1, characterized in that: The improved differential evolution algorithm based on the random adaptive operator in step S5 includes the following sub-steps: S5.1 Initialize the population The population is defined as: , where D is the dimension of the solution space and the vector Indicates The "chromosome" of the generation population. Indicates Daidi Chromosome No. Genes: In the formula, and Respectively represent chromosome The upper and lower bounds of the gene value range, . Represents the size of the population Is the interval The random number in . S5.2 Mutation Operation The differential evolution algorithm uses the differential strategy to achieve individual mutation, that is, randomly select two different individuals in the population, scale their vector difference, and synthesize the vector with the individual to be mutated to generate an intermediate individual. : In the formula, is the best individual in the current population, is a random selection of individuals in the population, and . Random adaptive scaling factor It is a scaling factor whose value fluctuates and decreases as the number of iterations increases. The formula is as follows: In the formula, , represents the scaling factor The initial value of . represents the maximum number of iterations, Indicates the current number of iterations, function Represents the mean and standard deviation Generate random numbers from a normal distribution, generally , The value is relatively small to describe small random fluctuations and prevent the differential evolution algorithm from falling into a local optimal solution. S5.3: Crossover Operations For Generation population and its variant intermediates Perform crossover operations between individuals: In the formula, For interval A random positive integer in is the crossover probability. S5.4: Boundary constraints In the process of evolution, in order to ensure the validity of the solution, it is necessary to determine whether each "gene" in the "chromosome" satisfies the constrained boundary conditions. If the range of population individuals exceeds the given upper and lower limits, the boundary absorption method is used for boundary constraints: S5.5: Select an action According to the greedy criterion, individuals entering the next generation population are selected: In the formula, For vector The fitness function is defined as follows: In the formula, is the daily transaction cost of the router. The specific expression is shown in S4.1 Power router transaction cost model. It is the power distribution of power trading and load reduction of power routers. S5.6: Based on the optimization results, an internal optimization scheduling strategy for the power router is proposed to provide reference values ​​for power trading between the power router and the distribution network and the power distribution of the load that can be reduced.

7. The energy coordination control method of an electric energy router according to claim 2, characterized in that: In step S2.2, the BiLSTM neural network prediction model based on the sparrow algorithm can be divided into the following sub-steps: S6.1 Sparrow Optimization Algorithm: The sparrow population is divided into discoverers, followers, and scouts. The discoverers are responsible for exploring new food sources and guiding the search direction of the group; followers search for food near the discoverers and conduct local searches; scouts observe whether there is any danger inside the sparrow group and remind the entire sparrow group to be safe. Sparrow Collection Matrix as follows: Where: , It is the size of a sparrow. . The fitness value of all sparrows is Where: , Each value in represents the fitness value of an individual, is the dimension of the variable. It indicates a sparrow with a better fitness value, and this sparrow is the first to obtain food and, as a discoverer, leads the entire population toward the food source. set up is a small positive number approaching zero, then the basis for judging whether the sparrow has reached the food source is: The position update formula of each generation of discoverers is as follows: Where: is the pre-set maximum number of iterations, for A uniform random number in is a standard normally distributed random number. The position of each follower generation is updated as follows: Where: It is the sparrow with the worst fitness in the current population. ,in A row matrix where all elements are randomly set to 1 or -1. Scouts need to keep an eye on the current environment so that they can immediately fly to a safe environment when they find danger. The position update is as follows: Where: is the sparrow with the best fitness in the current population, is a random number that conforms to the normal distribution. for A random number between to a smaller number to prevent the denominator from being 0. is the worst fitness value of the current population. is the global optimal fitness value. S6.2 BiLSTM neural network prediction model: The single-layer BiLSTM in BiLSTM is composed of two continuous time step LSTMs, namely the forward continuous time step LSTM and the reverse continuous time step LSTM. Its core structure is a single-step LSTM network model, which includes three sigmoid activation functions. Keep the output range between 0 and 1. Each sigmoid activation function belongs to a flexible gate function. The gate function is divided into three different gate functions: forget gate , input gate , output gate ; a hyperbolic tangent function ; Input data timing , output data timing The three gate functions decide whether to forget or retain some or all of the inherited data through the current input and the output of the previous time step. The weight values ​​are learned through the back propagation algorithm. and deviation Narrow the gap between the predicted output and the actual training sample. Through this forget-retain-output iterative process, the current information can be retained, and the output is concatenated after forward and backward processing of the input sequence, and the final result is output. The functional relationship of the single time step LSTM network prediction model is as follows: in , , , , , , and is the weight matrix of each LSTM network module. , , and is the bias in the LSTM network. represents the sigmoid startup function, Represents the hyperbolic tangent function. S6.3 Optimal output vector according to the sparrow algorithm , set the hyperparameters of the BiLSTM neural network prediction model, such as the learning rate, number of hidden layers, and maximum training rounds, and predict the real-time output of each unit of the power router based on historical data.

8. The energy coordination control method of an electric energy router according to claim 3 is characterized by: In step S3.3, the transferable load scheduling method based on time-of-use electricity price can be divided into the following sub-steps: S7.1 The cost objective function of the transferable load based on time-of-use electricity price is: S7.2 The constraints are: (1) The transfer power satisfies the constraints: in, and Respectively i The upper and lower limits of the transfer capacity at all times. (2) The minimum continuous operation time constraint of the transferable load is: S.7.3 According to the improved differential evolution algorithm based on random adaptive operators as described in claim 5, electricity consumption is regulated based on time-of-use electricity prices. Under the conditions of the allowable electricity consumption time of the transferable load, the load output corresponding to the electricity consumption time period of the transferable load can be obtained, so as to minimize the electricity cost while meeting the electricity demand.

Images