Microgrid operation optimization method based on cost-sensitive decision tree

Abstract

The invention relates to a micro-grid operation optimization method based on a cost-sensitive decision tree. The micro-grid operation optimization method comprises the following steps of S1, performing data acquisition of equipment parameters, load parameters and natural condition parameters; s2, constructing an unbalanced data classification model taking maximization of information gain as a target; s3, introducing a rejection cost optimization mechanism, and updating the unbalanced data classification model; s4, completing the construction of a micro-grid operation optimization model; and S5, obtaining an optimal operation scheduling scheme. According to the method, the optimized operation level of the micro-grid can be remarkably improved, and more efficient and reliable energy management is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of microgrid operation optimization, and relates to a microgrid operation optimization method, in particular to a microgrid operation optimization method based on a cost-sensitive decision tree. Background Art

[0002] With the progress of innovation in the power market and the increasing demand for versatility from users, participation in demand-side response is an important means to achieve coordinated operation of "source-grid-load-storage". However, with the continuous expansion of the scale of renewable energy systems (RES), problems such as insufficient peak-reduction capacity of the power grid and limited unit output have become increasingly prominent. In order to further coordinate random factors, it is necessary to study the internal micro-source and energy storage scheduling of microgrids. Strengthening the use and cost management of various energy loads on the demand side can not only improve the matching degree of multi-energy supply and demand, but also encourage users to change their energy consumption habits, which is conducive to the energy saving, efficiency and development of microgrids.

[0003] In existing research, various methods have been proposed for the optimal operation optimization of microgrids. In the latest research, some scholars have used compressed air energy storage as one of the important energy storage technologies to assist wind farms in regulating peak loads and providing other grid auxiliary services. This method reduces the overall dispatching cost and efficiency in smoothing the output of the unit. However, the above methods still have the following defects: they fail to fully consider the efficiency fluctuations of different energy storage methods of the energy storage system, resulting in instability in the regulation capacity. More importantly, the lack of a comprehensive assessment of the overall reliability and flexibility of the system limits the dispatching capacity and response speed of the microgrid, and thus fails to achieve the expected optimization effect. Therefore, how to comprehensively consider different types of loads and their characteristics to achieve mutual improvement of power supply and demand is still a long-term challenge facing the optimization of microgrid operation.

[0004] After searching, no published technical documents identical or similar to the present invention were found. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and propose a microgrid operation optimization method based on a cost-sensitive decision tree. By analyzing the operating characteristics and mathematical models of distributed energy, combined with the unbalanced data in the operation of the microgrid, a cost-sensitive decision tree is used for classification and optimization, and a microgrid optimization scheduling plan is obtained, which can significantly improve the optimized operation level of the microgrid and achieve more efficient and reliable energy management.

[0006] The present invention solves the practical problem by adopting the following technical solutions:

[0007] A microgrid operation optimization method based on a cost-sensitive decision tree comprises the following steps:

[0008] Step S1: Conduct an in-depth analysis of the distributed energy operation characteristics in the microgrid, and construct a microgrid system model including a wind and solar power consumption and output model and a CCHP system efficiency and output model; based on the constructed model, collect data on equipment parameters, load parameters and natural condition parameters.

[0009] Step S2: Based on the unbalanced data collected in step 1, a cost-sensitive reinforcement learning algorithm is introduced to establish an unbalanced data classification model. The model uses the unbalanced data as the input of the decision tree algorithm, and sets a higher misclassification cost for minority class samples with higher output efficiency and smaller heat dissipation loss coefficient. An unbalanced data classification model is constructed with the goal of maximizing information gain, and then the efficiency and power under different operating conditions are classified.

[0010] Step S3: introducing a rejection cost optimization mechanism, optimizing the rejection cost, enhancing the sensitivity of cost learning to classification errors, and updating the imbalanced data classification model;

[0011] Step S4, combining the wind and solar power consumption and output model of the microgrid in step 1, the CCHP efficiency and output model and the characteristics of the microgrid itself, to establish an operation optimization model for the microgrid operation; construct an objective function through energy storage energy flow, comprehensive energy cost and the minimum rejection cost in step 3, and construct constraint conditions according to the prediction errors of wind power, photovoltaic power, electricity, cooling and heat in the microgrid, and complete the construction of the microgrid operation optimization model;

[0012] Step S5: Based on the objective function and microgrid probability constraints in step 4, an optimized decision tree algorithm is used to solve and obtain an optimal operation scheduling plan.

[0013] Moreover, the specific steps of step S1 include:

[0014] S1.1 Analyze the operating characteristics of distributed energy in microgrids and build a wind and solar energy consumption and output model for microgrid systems, including the output power of wind and solar energy;

[0015] The calculation formula of wind power output PWT is:

[0016]

[0017] Where ρ is the current air density (kg / m3), v is the wind speed (m / s), R WT is the radius of the wind turbine (m), C p is the power coefficient in the range of 0.35-0.45, A is the rotor swept area (m 2 ), N g is the generator efficiency, N b is the gearbox bearing efficiency;

[0018] The output power of the photovoltaic panel is expressed as:

[0019]

[0020] Where P pv is the output power of the photovoltaic cell in the t period; P STC and I STC is the photovoltaic power and current under standard test conditions; K is the power coefficient at a specific temperature; T e (t) and T ref are the temperature and reference temperature values ​​in the tth period respectively.

[0021] S1.2, construct the CCHP system efficiency and output model. The CCHP system includes a production capacity part and an energy storage part. The production capacity part includes: gas turbine, absorption refrigerator, gas boiler, and fuel cell.

[0022] The efficiency and output model of the production capacity section includes:

[0023] The mathematical model of the micro gas turbine's fuel consumption and its natural gas consumption is expressed as:

[0024]

[0025] Where U MT (t) is the fuel consumption of the micro gas turbine in period t; P MT (t) and η MT (t) are the output power and efficiency of the micro gas turbine in period t; Δt is the unit scheduling time; L NG It is natural gas with a lower calorific value;

[0026] The combined heat and power system realizes cooling and heating under the drive of electrical energy, making T CMT and T HMT are the cooling and heating powers of the cogeneration system at time t, respectively. The expressions of cooling, heating and electric energy output power are:

[0027]

[0028] T CMT (t) = η rec R MTC T MT (t)(7)

[0029] T HMT (t) = η rec R MTH T MT (t)(8)

[0030] Where η MTis the heat rejection coefficient of the micro-turbine in the tth period; η L is the heat loss coefficient; η ref is the gas flow coefficient; R MTC and R MTH They are the cooling coefficient and heating coefficient of the absorption chiller and waste heat recovery unit respectively.

[0031] Fuel cell efficiency and output power are expressed as:

[0032] η FC =0.6744(P FC / I FC ) (9)

[0033] Where FC is the fuel cell; η FC is the fuel cell efficiency; P FC is the fuel cell power (W), I FC is the current of the fuel cell (A).

[0034] The energy storage part of CCHP is modeled. The mathematical model of the energy storage system that meets the energy charging and discharging constraints, energy storage capacity constraints and other conditions during operation and has relatively similar operating characteristics can be expressed as:

[0035]

[0036]

[0037] Where P B is the power of the energy storage unit in the tth period, -P B,ch (t), +P B,dis (t) are the charging and discharging power of the battery energy storage unit in the tth period.

[0038] S1.3 Collect equipment parameters, load parameters and natural condition parameters of wind energy, solar energy and natural gas, electricity, cooling energy and thermal energy sources, temperature and power in photovoltaics, and energy efficiency of each component in CCHP.

[0039] Moreover, the specific steps of step 2 include:

[0040] S2.1 Based on the unbalanced data collected in step 1, the cost-sensitive reinforcement learning algorithm is introduced to establish an unbalanced data classification model. When dealing with the microgrid parameter classification problem, <x i ,y i > represents the input equipment, load and natural condition parameter samples, x i Represents sample attributes, y i ∈{1,2,…,N} represents the sample type, P(j|x) is the posterior probability of sample x, C ijIt represents the cost of misclassifying the i-th class sample into the j-th class. R(x,i) represents the misclassification cost of classifying the sample x into class i, which is calculated by formula (12):

[0041]

[0042] Through calculation, the category with the smallest R(x,i) is selected as the final classification result:

[0043]

[0044] For the classification problem of balanced data, it is equivalent to setting the cost of misclassification of different categories to 1. The classification cost at this time becomes:

[0045]

[0046] This formula assigns a higher misclassification cost to the minority class samples corresponding to the cases with higher output efficiency and smaller heat dissipation loss coefficient.

[0047] S2.2 sets the initial value of the decision tree algorithm, and takes the unbalanced data such as air density, wind speed, wind turbine radius, temperature and power in photovoltaic, and energy efficiency of each component in CCHP preprocessed in step 1 as input; use the array x i represents the i-th group of known equipment, loads and natural conditions, y i Represents the corresponding maximum information gain IG, and its expression is:

[0048]

[0049] Among them, A is the classification feature, Values(A) is the set of its feature values, and x v represents the subset corresponding to the feature A when its value is v; as shown in formula (16), H(x i ) is the data set x i The entropy measures the uncertainty or purity of samples in a data set. The higher the entropy, the more uneven the distribution of sample categories and the greater the difficulty of classification. i is the proportion of category i in the data set; classification is performed with the maximum information gain feature maxIG as the target, using 1 and 0 to represent true and false respectively; when the maximum information gain condition is met, the classification model of unbalanced data is constructed, and then the efficiency and power under different operating conditions are classified to identify operation optimization opportunities.

[0050] Moreover, the specific steps of step 3 include:

[0051] The rejection cost is introduced to optimize the classification. The rejection function f(t) and the threshold δ are defined. If f(t)>δ, the sample is classified normally; if f(t)<δ, the rejection strategy is adopted and the sample is not classified. Taking the binary classification problem as an example, for the sample <x i ,y i >, the final classification result is shown in formula (17):

[0052]

[0053] Among them, y opt Indicates the final class label of the sample, n means that the sample is divided into n categories, p means that the sample is divided into p categories, and 0 means that the sample is not classified;

[0054] For samples <x i ,y i >, update the mixed cost matrix C ij for:

[0055] C ij ={C ij (p,n),C ij (n,p),C ij (0,n),C ij (0,p)}(18)

[0056] Among them C ij (0,n) and C ij (0,p) represents the rejection cost. The final classification result is updated with the minimum rejection cost as the goal:

[0057] y opt =argmin{R(i|t)}=argmin{∑P(j|x)·C(i,j)},i,j∈{0,n,p}(19)

[0058] Replace formula (13) with formula (19) and proceed to step S2.2 to complete the classification, update the imbalanced data classification model, and improve the overall classification performance.

[0059] Moreover, the specific steps of step 4 include:

[0060] S4.1 establishes the objective function of the energy storage energy flow of the optimization operation model. The microgrid system model and energy flow process constructed according to the architecture analyzed by the energy storage model, taking into account the timing expressions of the electric, cold and hot loads after controllable energy storage in the microgrid system are as follows:

[0061]

[0062] Among them, load, controllable load, Q c (t) and Q h(t) are electric load, cooling load and heating load respectively, * represents the corresponding parameter after t cycle; V p,ch (t), V C,ch (t) and V H,ch (t), electricity, cold, heat, energy shortage state variable is 1, in energy shortage state; P p,ch (t), Q C,ch (t) and Q H,ch (t) are the power shortage rates of electricity, cooling and heating energy in time period t; V P,dis (t), V C,dis (t), V H,dis (t) is the state variable of excess energy of electricity, cold and heat; P P,dis (t), Q C,dis (t), Q H,dis (t) is the discharge power rate of electric refrigeration and thermal energy.

[0063] S4.2 Considering the actual energy consumption cost, a comprehensive energy cost objective function of the microgrid system is established, including unit power generation cost (unit fuel cost, operation and maintenance cost, unit startup cost), demand response cost, and cooling (heating) benefits of the cogeneration system.

[0064] The comprehensive operating cost function of the system is:

[0065]

[0066] in

[0067]

[0068] C CH (t) = K c [Q c (t)-Q C,ch (t)]+K h [Q he (t)-Q H,ch (t)](27)

[0069] In the formula, F G is the comprehensive operating cost of the system; C F is the fuel cost, C D (t) is the demand cost; C M is the maintenance cost, C CH (t) is the sum of the cooling (heating) profits of the cogeneration system in each time period t.

[0070] S4.3 constructs the minimum rejection cost objective function. Combined with step 3, the final classification result is updated with the minimum rejection cost as the goal as shown in formula (19). R(i|t) represents higher output efficiency and smaller heat dissipation loss coefficient.

[0071] S4.4 adds microgrid constraints. Assuming that the prediction errors of wind power, photovoltaic power, electricity, cooling and heat follow a normal distribution, the microgrid probability constraint is expressed as

[0072]

[0073] Where P i CG is the power of a conventional generator; is the power of the renewable generator; Q IS For cold energy storage; Q TS is the thermal energy storage; α, β and γ are standard normal distributions.

[0074] Moreover, the specific steps of step 5 include:

[0075] S5.1 sets the initial value of the decision tree algorithm, takes the energy storage energy flow, comprehensive energy cost and minimum rejection cost classification objective function in step 4 as different features, calculates the information gain of these features, selects the i-th decision tree for the i-th output efficiency and heat dissipation loss coefficient, and selects the feature with the largest information gain as the best feature for data partitioning after multiple selections to generate the data set x min .

[0076] S5.2 sets the decision stopping condition, i.e., the constraint condition in step 4. For each partitioned subset x min , check the stopping condition. If the stopping condition is not met, recursively repeat the feature selection and optimizable node partitioning process for each subset to build a deep structure of the decision tree until all subsets meet the conditions.

[0077] S5.3 After the decision tree is constructed, the optimization decision is marked for each optimizable node, and the optimal decision is selected as the output of the node. In addition, the pruning step combines the operation efficiency of the microgrid, calculates the decision effect of each node, performs pre-pruning or post-pruning, removes redundant nodes, and obtains the optimal operation plan.

[0078] Advantages and beneficial effects of the present invention:

[0079] 1. The present invention proposes a microgrid operation optimization method based on a cost-sensitive decision tree. First, the basic structure of the microgrid is focused on, the main wind and solar power consumption and output model and CCHP efficiency and output model of the microgrid system are analyzed, and the unbalanced data are pre-processed. Then, we input these unbalanced data into the pre-built cost-sensitive learning model and introduce the decision tree algorithm. Subsequently, the optimal operation objective function and conditional constraints are updated by combining the initially constructed microgrid system wind and solar power consumption and output model, CCHP efficiency and output model and actual consumption data, and finally the optimal operation plan is obtained.

[0080] 2. When constructing a microgrid model, step 1 of the present invention not only considers basic wind energy, solar energy and natural gas, but also considers the energy storage efficiency of electric energy, cooling energy and thermal energy, which helps to ensure the stability and predictability of system construction.

[0081] 3. The cost-sensitive reinforcement learning algorithm is introduced in step 2 of the present invention. The algorithm takes unbalanced data as input, assigns a higher misclassification cost to minority class samples, and performs classification with the goal of maximizing information gain. The core of this stage is to improve the algorithm's recognition ability for minority classes through a cost-sensitive mechanism, thereby improving the overall learning performance.

[0082] 4. The present invention uses a method combining cost-sensitive learning and decision trees. For example, in steps 3-5, we use classification cost to select the splitting attribute of the decision tree to enhance the accuracy of classification. The decision tree also conducts a comprehensive evaluation of the overall reliability and flexibility of the system step by step, thereby improving the operation and dispatching capability and response speed of the microgrid. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A processing flow chart of a microgrid operation optimization method based on a cost-sensitive decision tree of the present invention;

[0084] Figure 2 It is a schematic diagram of the structure and energy flow of the microgrid;

[0085] Figure 3 It is a schematic diagram of the construction of a proton conducting fuel cell;

[0086] Figure 4 It is the flow chart of cost-sensitive decision tree algorithm. DETAILED DESCRIPTION

[0087] The embodiments of the present invention are further described in detail below with reference to the accompanying drawings:

[0088] A microgrid operation optimization method based on cost-sensitive decision tree, such as Figure 1 As shown, the following steps are included:

[0089] Step S1: Conduct an in-depth analysis of the distributed energy operation characteristics in the microgrid, and construct a microgrid system model including a wind and solar power consumption and output model and a CCHP system efficiency and output model; based on the constructed model, collect data on equipment parameters, load parameters and natural condition parameters.

[0090] The specific steps of step S1 include:

[0091] S1.1 Analyze the operating characteristics of distributed energy in microgrids. The microgrid structure and energy flow process are as follows: Figure 2As shown, a wind and solar energy consumption and output model of the microgrid system is constructed, including the output power of wind and solar energy;

[0092] Wind power output (PWT) is mainly determined by wind speed and utilization rate, and its calculation formula is:

[0093]

[0094] Where ρ is the current air density (kg / m3), v is the wind speed (m / s), R WT is the radius of the wind turbine (m), C p is the power coefficient in the range of 0.35-0.45, A is the rotor swept area (m 2 ), N g is the generator efficiency, N b is the gearbox bearing efficiency.

[0095] The output power of photovoltaic panels varies with the changes in external temperature and light intensity. It is obtained by using the power value of standard test conditions as a reference. The expression of output power is:

[0096]

[0097] Where P pv is the output power of the photovoltaic cell in the t period; P STC and I STC is the photovoltaic power and current under standard test conditions; K is the power coefficient at a specific temperature; T e (t) and T ref are the temperature and reference temperature values ​​in the tth period respectively.

[0098] S1.2, construct the CCHP system efficiency and output model. The CCHP system includes a production capacity part and an energy storage part. The production capacity part includes: gas turbine, absorption refrigerator, gas boiler, and fuel cell.

[0099] The efficiency and output model of the production capacity section includes:

[0100] The internal structure of a micro gas turbine Figure 3 As shown, the mathematical model of its fuel consumption and natural gas consumption is expressed as:

[0101]

[0102] Where U MT (t) is the fuel consumption of the micro gas turbine in period t; P MT (t) and η MT (t) are the output power and efficiency of the micro gas turbine in period t; Δt is the unit scheduling time; L NGIt is natural gas with a lower calorific value;

[0103] The combined heat and power system realizes cooling and heating under the drive of electrical energy, making T CMT and T HMT are the cooling and heating powers of the cogeneration system at time t, respectively. The expressions of cooling, heating and electric energy output power are:

[0104]

[0105] T CMT (t) = η rec R MTC T MT (t)(7)

[0106] T HMT (t) = η rec R MTH T MT (t)(8)

[0107] Where η MT is the heat rejection coefficient of the micro-turbine in the tth period; η L is the heat loss coefficient; η ref is the gas flow coefficient; R MTC and R MTH They are the cooling coefficient and heating coefficient of the absorption chiller and waste heat recovery unit respectively.

[0108] Fuel cell efficiency and output power can be expressed as:

[0109] η FC =0.6744(P FC / I FC )(9)

[0110] Where FC is the fuel cell; η FC is the fuel cell efficiency; P FC is the fuel cell power (W), I FC is the current of the fuel cell (A).

[0111] The energy storage part of CCHP is modeled. The mathematical model of the energy storage system that meets the energy charging and discharging constraints, energy storage capacity constraints and other conditions during operation and has relatively similar operating characteristics can be expressed as

[0112]

[0113] Where P B is the power of the energy storage unit in the tth period, -P B,ch (t), +P B,dis (t) are the charging and discharging power of the battery energy storage unit in the tth period.

[0114] S1.3 Collect equipment parameters, load parameters and natural condition parameters of wind energy, solar energy and natural gas, electricity, cooling energy and thermal energy sources, such as air density, wind speed (m / s), wind turbine radius (m) in the wind turbine, temperature and power in photovoltaics, and energy efficiency of each component in CCHP.

[0115] In step S2, based on the unbalanced data collected in step 1, a cost-sensitive reinforcement learning algorithm is introduced to establish an unbalanced data classification model. The model uses the unbalanced data as the input of the decision tree algorithm, and sets a higher misclassification cost for minority class samples with higher output efficiency and smaller heat dissipation loss coefficient. An unbalanced data classification model is constructed with the goal of maximizing information gain, and then the efficiency and power under different operating states are classified to identify operation optimization opportunities.

[0116] The specific steps of step 2 include:

[0117] S2.1 Based on the unbalanced data collected in step 1, the cost-sensitive reinforcement learning algorithm is introduced to establish an unbalanced data classification model. When dealing with the microgrid parameter classification problem, <x i ,y i > represents the input equipment, load and natural condition parameter samples, x i Represents sample attributes, y i ∈{1,2,…,N} represents the sample type, P(j|x) is the posterior probability of sample x, C ij It represents the cost of misclassifying the i-th class sample into the j-th class. R(x,i) represents the misclassification cost of classifying the sample x into class i, which can be calculated by formula (12):

[0118]

[0119] Through calculation, the category with the smallest R(x,i) is selected as the final classification result:

[0120]

[0121] For the classification problem of balanced data, it is equivalent to setting the cost of misclassification of different categories to 1. The classification cost at this time becomes:

[0122]

[0123] This formula can assign a higher misclassification cost to minority class samples corresponding to cases with higher output efficiency and smaller heat dissipation loss coefficient.

[0124] S2.2 sets the initial value of the decision tree algorithm, and takes the unbalanced data such as air density, wind speed (m / s), wind turbine radius (m), temperature and power in photovoltaic, and energy efficiency of each component in CCHP preprocessed in step 1 as input. i represents the i-th group of known equipment, loads and natural conditions, y i Represents the corresponding maximum information gain (InformationGain, IG), and its expression is:

[0125]

[0126] Among them, A is the classification feature, Values(A) is the set of its feature values, and x v represents the subset corresponding to the feature A when its value is v. As shown in formula (16), H(x i ) is the data set x i The entropy measures the uncertainty or purity of samples in a data set. The higher the entropy, the more uneven the distribution of sample categories and the greater the difficulty of classification. i is the proportion of category i in the data set. Classification is performed with the maximum information gain feature maxIG as the target, using 1 and 0 to represent true and false respectively; when the maximum information gain condition is met, the classification model of unbalanced data is constructed, and then the efficiency and power under different operating conditions are classified to identify operation optimization opportunities.

[0127] Step S3, in order to further improve the classification accuracy of the imbalanced data classification model established in step 2 with the goal of maximizing information gain, a rejection cost optimization mechanism is introduced. By optimizing the rejection cost, the sensitivity of cost learning to classification errors is enhanced, the imbalanced data classification model is updated, and the overall classification performance is improved.

[0128] The specific steps of step 3 include:

[0129] Cost-sensitive learning (CSL) can make up for the one-sidedness of the classifier that only focuses on improving the classification accuracy. Here, the rejection cost is selected to identify the samples.

[0130] The rejection cost is introduced to optimize the classification. The rejection function f(t) and the threshold δ are defined. If f(t)>δ, the sample is classified normally; if f(t)<δ, the rejection strategy is adopted and the sample is not classified. Taking the binary classification problem as an example, for the sample <x i ,y i >, the final classification result is shown in formula (17):

[0131]

[0132] Here y optRepresents the final class label of the sample, n means that the sample is divided into n categories, p means that the sample is divided into p categories, and 0 means that the sample is not classified.

[0133] For samples <x i ,y i >, update the mixed cost matrix C ij for:

[0134] C ij ={C ij (p,n),C ij (n,p),C ij (0,n),C ij (0,p)}(18)

[0135] Among them C ij (0,n) and C ij (0,p) represents the rejection cost. The final classification result is updated with the minimum rejection cost as the goal:

[0136] y opt =argmin{R(i|t)}=argmin{∑P(j|x)·C(i,j)},i,j∈{0,n,p}(19)

[0137] Replace formula (13) with formula (19) and proceed to step S2.2 to complete the classification, update the imbalanced data classification model, and improve the overall classification performance.

[0138] Step S4, combining the wind and solar power consumption and output model of the microgrid in step 1, the CCHP efficiency and output model and the characteristics of the microgrid itself, establish an operation optimization model for the microgrid operation; construct the objective function through the energy storage energy flow, the comprehensive energy cost and the minimum rejection cost in step 3, and construct constraint conditions according to the prediction errors of wind power, photovoltaic power, electricity, cooling and heat in the microgrid to complete the construction of the microgrid operation optimization model.

[0139] The specific steps of step 4 include:

[0140] S4.1 Establish the objective function of the energy storage energy flow of the optimized operation model according to Figure 2 The microgrid system model and energy flow process constructed by the architecture of the energy storage model analysis shown in the figure take into account the timing expressions of the electric, cold and hot loads after the controllable energy storage in the microgrid system as follows:

[0141]

[0142] Among them, load, controllable load, Q c (t) and Q h (t) are electric load, cooling load and heating load respectively, * represents the corresponding parameter after t cycle; Vp,ch (t), V C,ch (t) and V H,ch (t), electricity, cold, heat, energy shortage state variable is 1, in energy shortage state; P p,ch (t), Q C,ch (t) and Q H,ch (t) are the power shortage rates of electricity, cooling and heating energy in time period t; V P,dis (t), V C,dis (t), V H,dis (t) is the state variable of excess energy of electricity, cold and heat; P P,dis (t), Q C,dis (t), Q H,dis (t) is the discharge power rate of electric refrigeration and thermal energy.

[0143] S4.2 Considering the actual energy consumption cost, a comprehensive energy cost objective function of the microgrid system is established, including unit power generation cost (unit fuel cost, operation and maintenance cost, unit startup cost), demand response cost, and cooling (heating) benefits of the cogeneration system.

[0144] The comprehensive operating cost function of the system is:

[0145]

[0146] in

[0147]

[0148] C CH (t) = K c [Q c (t)-Q C,ch (t)]+K h [Q he (t)-Q H,ch (t)](27)

[0149] In the formula, F G is the comprehensive operating cost of the system; C F is the fuel cost, C D (t) is the demand cost; C M is the maintenance cost, C CH (t) is the sum of the cooling (heating) profits of the cogeneration system in each time period t.

[0150] S4.3 constructs the minimum rejection cost objective function. Combined with step 3, the final classification result is updated with the minimum rejection cost as the goal as shown in formula (19). R(i|t) represents higher output efficiency and smaller heat dissipation loss coefficient.

[0151] S4.4 adds microgrid constraints. Assuming that the prediction errors of wind power, photovoltaic power, electricity, cooling and heat follow a normal distribution, the microgrid probability constraint is expressed as

[0152]

[0153] Where P i CG is the power of a conventional generator; is the power of the renewable generator; Q IS For cold energy storage; Q TS is the thermal energy storage; α, β and γ are standard normal distributions.

[0154] Step S5, based on the objective function in step 4 and the probability constraint of the microgrid, an optimized decision tree algorithm is used to solve the problem and complete the construction of the optimal operation and dispatching method of the microgrid.

[0155] The specific steps of step 5 include:

[0156] S5.1 sets the initial value of the decision tree algorithm, takes the energy storage energy flow, comprehensive energy cost and minimum rejection cost classification objective function in step 4 as different features, calculates the information gain of these features, selects the i-th decision tree for the i-th output efficiency and heat dissipation loss coefficient, and selects the feature with the largest information gain as the best feature for data partitioning after multiple selections to generate the data set x min .

[0157] S5.2 sets the decision stopping condition, i.e., the constraint condition in step 4. For each partitioned subset x min , check the stopping condition. If the stopping condition is not met, recursively repeat the feature selection and optimizable node partitioning process for each subset to build a deep structure of the decision tree until all subsets meet the conditions.

[0158] S5.3 After the decision tree is constructed, mark the optimization decision for each optimizable node and select the best decision as the output of the node. In addition, the pruning step combines the operating efficiency of the microgrid, calculates the decision effect of each node, performs pre-pruning or post-pruning, removes redundant nodes, ensures the robustness and effectiveness of the model under complex operating conditions, and improves the adaptability and performance of the decision tree in the actual application of the microgrid. The optimal operation scheduling plan is obtained. The specific process is as follows Figure 4 shown.

[0159] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0160] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0161] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0162] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

Claims

1. A microgrid operation optimization method based on a cost-sensitive decision tree, characterized in that: The following steps are involved: Step S1: Conduct an in-depth analysis of the distributed energy operation characteristics in the microgrid, and construct a microgrid system model including a wind and solar power consumption and output model and a CCHP system efficiency and output model; based on the constructed model, collect data on equipment parameters, load parameters, and natural condition parameters; Step S2: Based on the unbalanced data collected in step 1, a cost-sensitive reinforcement learning algorithm is introduced to establish an unbalanced data classification model. The model uses the unbalanced data as the input of the decision tree algorithm, and sets a higher misclassification cost for minority class samples with higher output efficiency and smaller heat dissipation loss coefficient, thereby constructing an unbalanced data classification model with the goal of maximizing information gain, and then classifying the efficiency and power under different operating states; Step S3: introducing a rejection cost optimization mechanism, optimizing the rejection cost, enhancing the sensitivity of cost learning to classification errors, and updating the imbalanced data classification model; Step S4, combining the wind and solar power consumption and output model of the microgrid in step 1, the CCHP efficiency and output model and the characteristics of the microgrid itself, to establish an operation optimization model for the microgrid operation; construct an objective function through energy storage energy flow, comprehensive energy cost and the minimum rejection cost in step 3, and construct constraint conditions according to the prediction errors of wind power, photovoltaic power, electricity, cooling and heat in the microgrid, and complete the construction of the microgrid operation optimization model; Step S5: Based on the objective function and microgrid probability constraints in step 4, an optimized decision tree algorithm is used to solve and obtain an optimal operation scheduling plan.

2. The microgrid operation optimization method based on a cost-sensitive decision tree according to claim 1, characterized in that: The specific steps of step S1 include: S1.1 Analyze the operating characteristics of distributed energy in microgrids and build a wind and solar energy consumption and output model for microgrid systems, including the output power of wind and solar energy; The calculation formula of wind power output PWT is: Where ρ is the current air density (kg / m3), v is the wind speed (m / s), R WT is the radius of the wind turbine (m), C p is the power coefficient in the range of 0.35-0.45, A is the rotor swept area (m 2 ), N g is the generator efficiency, N b is the gearbox bearing efficiency; The output power of the photovoltaic panel is expressed as: Where P pv is the output power of the photovoltaic cell in the t period; P STC and I STC is the photovoltaic power and current under standard test conditions; K is the power coefficient at a specific temperature; T e (t) and T ref are the temperature and reference temperature value of the tth period respectively; S1.2, construct the CCHP system efficiency and output model. The CCHP system includes a production capacity part and an energy storage part. The production capacity part includes: gas turbine, absorption refrigerator, gas boiler, and fuel cell. The efficiency and output model of the production capacity section includes: The mathematical model of the micro gas turbine's fuel consumption and its natural gas consumption is expressed as: Where U MT (t) is the fuel consumption of the micro gas turbine in period t; P MT (t) and η MT (t) are the output power and efficiency of the micro gas turbine in period t; Δt is the unit scheduling time; L NG It is natural gas with lower calorific value; The combined heat and power system realizes cooling and heating under the drive of electrical energy, making T CMT and T HMT are the cooling and heating powers of the cogeneration system at time t, respectively. The expressions of cooling, heating and electric energy output power are: T CMT (t)=η rec R MTC T MT (t) (7) T HMT (t)=η rec R MTH T MT (t) (8) Where η MT is the heat rejection coefficient of the micro-turbine in the tth period; η L is the heat loss coefficient; η ref is the gas flow coefficient; R MTC and R MTH are the cooling coefficient and heating coefficient of the absorption chiller and waste heat recovery unit respectively; Fuel cell efficiency and output power are expressed as: or FC =0.6744(P FC / I FC ) (9) Where FC is the fuel cell; η FC is the fuel cell efficiency; P FC is the fuel cell power (W), I FC is the current of the fuel cell (A); The energy storage part of CCHP is modeled. The mathematical model of the energy storage system that meets the energy charging and discharging constraints, energy storage capacity constraints and other conditions during operation and has relatively similar operating characteristics can be expressed as: Where P B is the power of the energy storage unit in the tth period, -P B,ch (t), +P B,dis (t) are the charging and discharging power of the battery energy storage unit in the tth period respectively; S1.3 Collect equipment parameters, load parameters and natural condition parameters of wind energy, solar energy and natural gas, electricity, cooling energy and thermal energy sources, temperature and power in photovoltaics, and energy efficiency of each component in CCHP.

3. The microgrid operation optimization method based on cost-sensitive decision tree according to claim 1 is characterized in that: The specific steps of step 2 include: S2.1 Based on the unbalanced data collected in step 1, the cost-sensitive reinforcement learning algorithm is introduced to establish an unbalanced data classification model. When dealing with the microgrid parameter classification problem, <x i ,y i > represents the input equipment, load and natural condition parameter samples, x i Represents sample attributes, y i ∈{1,2,…,N} represents the sample type, P(j|x) is the posterior probability of sample x, C ij represents the cost of misclassifying the i-th class sample into the j-th class; R(x,i) represents the misclassification cost of classifying the sample x into class i, which is calculated by formula (12): Through calculation, the category with the smallest R(x,i) is selected as the final classification result: For the classification problem of balanced data, it is equivalent to setting the cost of misclassification of different categories to 1. The classification cost at this time becomes: This formula assigns a higher misclassification cost to the minority class samples corresponding to the cases with higher output efficiency and smaller heat dissipation loss coefficient; S2.2 sets the initial value of the decision tree algorithm, and takes the unbalanced data such as air density, wind speed, wind turbine radius, temperature and power in photovoltaic, and energy efficiency of each component in CCHP preprocessed in step 1 as input; use the array x i represents the i-th group of known equipment, loads and natural conditions, y i Represents the corresponding maximum information gain IG, and its expression is: Among them, A is the classification feature, Values(A) is the set of its feature values, and x v represents the subset corresponding to the feature A when its value is v; as shown in formula (16), H(x i ) is the data set x i The entropy measures the uncertainty or purity of samples in the data set. The higher the entropy, the more uneven the distribution of sample categories and the greater the difficulty of classification. i is the proportion of category i in the data set; classification is performed with the maximum information gain feature maxIG as the target, using 1 and 0 to represent true and false respectively; when the maximum information gain condition is met, the classification model of unbalanced data is constructed, and then the efficiency and power under different operating conditions are classified to identify operation optimization opportunities.

4. The microgrid operation optimization method based on a cost-sensitive decision tree according to claim 1, characterized in that: The specific steps of step 3 include: The rejection cost is introduced to optimize the classification. The rejection function f(t) and the threshold δ are defined. If f(t)>δ, the sample is classified normally; if f(t)<δ, a rejection strategy is adopted and the sample is not classified. Taking the binary classification problem as an example, for the sample <x i ,y i >, the final classification result is shown in formula (17): Among them, y opt Indicates the final class label of the sample, n means that the sample is divided into n categories, p means that the sample is divided into p categories, and 0 means that the sample is not classified; For samples <x i ,y i >, update the mixed cost matrix C ij for: C ij ={C ij (p,n),C ij (n,p),C ij (0,n),C ij (0,p)} (18) Among them C ij (0,n) and C ij (0, p) represents the rejection cost; the final classification result is updated with the minimum rejection cost as the goal: y opt =argmin{R(i|t)}=argmin{∑P(j|x)·C(i,j)}, i,j∈{0,n,p} (19) Replace formula (13) with formula (19) and proceed to step S2.2 to complete the classification, update the imbalanced data classification model, and improve the overall classification performance.

5. The microgrid operation optimization method based on cost-sensitive decision tree according to claim 1 is characterized in that: The specific steps of step 4 include: S4.1 establishes the objective function of the energy storage energy flow of the optimization operation model. The microgrid system model and energy flow process constructed according to the architecture analyzed by the energy storage model, taking into account the timing expressions of the electric, cold and hot loads after controllable energy storage in the microgrid system are as follows: Among them, load, controllable load, Q c (t) and Q h (t) are electric load, cooling load and heating load respectively, * represents the corresponding parameter after t cycle; V p,ch (t), V C,ch (t) and V H,ch (t), electricity, cold, heat, energy shortage state variable is 1, in energy shortage state; P p,ch (t), Q C,ch (t) and Q H,ch (t) are the power shortage rates of electricity, cooling and heating energy in time period t; V P,dis (t), V C,dis (t), V H,dis (t) is the state variable of excess energy of electricity, cold and heat; P P,dis (t), Q C,dis (t), Q H,dis (t) is the discharge power rate of electric refrigeration and thermal energy; S4.2 Considering the actual energy consumption cost, establish the comprehensive energy cost objective function of the microgrid system, including unit power generation cost (unit fuel cost, operation and maintenance cost, unit startup cost), demand response cost, and cooling (heating) benefits of the cogeneration system; The comprehensive operating cost function of the system is: in C CH (t)=K c [Q c (t)-Q C,ch (t)]+K h [Q he (t)-Q H,ch (t)] (27) In the formula, F G is the comprehensive operating cost of the system; C F is the fuel cost, C D (t) is the demand cost; C M is the maintenance cost, C CH (t) is the sum of the cooling (heating) profits of the cogeneration system in each time period t; S4.3 constructs the minimum rejection cost objective function, combined with step 3, and updates the final classification result with the minimum rejection cost as the goal as shown in formula (19), where R(i|t) represents higher output efficiency and smaller heat dissipation loss coefficient; S4.4 adds microgrid constraints; assuming that the prediction errors of wind power, photovoltaic power, electricity, cooling and heat follow a normal distribution, the microgrid probability constraint is expressed as in is the power of a conventional generator; is the power of the renewable generator; Q IS For cold energy storage; Q TS is the thermal energy storage; α, β and γ are standard normal distributions.

6. The microgrid operation optimization method based on cost-sensitive decision tree according to claim 1 is characterized in that: The specific steps of step 5 include: S5.1 sets the initial value of the decision tree algorithm, takes the energy storage energy flow, comprehensive energy cost and minimum rejection cost classification objective function in step 4 as different features, calculates the information gain of these features, selects the i-th decision tree for the i-th output efficiency and heat dissipation loss coefficient, and selects the feature with the largest information gain as the best feature for data partitioning after multiple selections to generate the data set x min ; S5.2 Set the decision stopping condition, i.e. the constraint condition in step 4; for each partitioned subset x min , check the stopping condition. If the stopping condition is not met, recursively repeat the feature selection and optimizable node partitioning process for each subset to build a deep structure of the decision tree until all subsets meet the conditions; S5.3 After the decision tree is constructed, the optimization decision is marked for each optimizable node, and the optimal decision is selected as the output of the node; in addition, the pruning step combines the operating efficiency of the microgrid, calculates the decision effect of each node, performs pre-pruning or post-pruning, removes redundant nodes, and obtains the optimal operating plan.

Images