JP2006048474A - Plant optimal operation planning device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a plant optimal operation planning device which efficiently makes an operation plan comprehensively considering minimization of plant operation cost, minimization of the amount of gas discharge, and minimization of maintenance cost. <P>SOLUTION: The plant optimal operation planning device includes a load estimating means 10, an optimizing means 20, and a regular plant simulator means 30 and plans the start/stop state and the amount of fuel injection of each plant component apparatus by the optimizing means 20 in consideration of minimization of plant operation cost, minimization of the amount of gas discharge, and minimization of maintenance cost and verifies inputs/outputs of respective plant component apparatus by the regular plant simulation means 30 and repeats these planning and verification to finally make the most suitable plant operation plan. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a plant optimum operation planning apparatus for minimizing plant operation costs, gas emissions, and maintenance costs for plant components in a plant that supplies energy such as electric power, heat, steam, and air. About.

  Various methods have been proposed in the past to determine the optimal operation of a plant using an optimization method after taking into account the start / stop of various plant components, but the input / output characteristics and operation patterns of plant components are generally Treat as a linear equation and formulate it as a mixed integer programming problem, or solve the problem, or consider the nonlinearity (including non-continuous, non-differentiable, etc.) and operation patterns of plant component equipment In general, a method for obtaining an optimum operation of a plant by a heuristic method capable of handling non-linearities such as expert systems and fuzzy reasoning has been generally used.

However, since the input / output characteristics of various plant components have nonlinearity and take different operation patterns depending on the time of day, it is usually difficult to formulate them as general equations so that they can be handled directly by optimization methods. It is. Therefore, the optimal operation of the plant needs to be formulated as a nonlinear mixed integer programming problem that is the most difficult to solve among the optimization problems.
However, after comprehensively considering discrete values such as start / stop of plant component equipment and continuous values such as fuel injection amount of plant component equipment, restrictions on characteristics of each plant component equipment and supply and demand for each load type Together with constraints such as balance and objective functions such as minimizing plant operating costs, exhaust gas minimization and maintenance costs, it is difficult to mathematically optimize for problems with non-linearities.

In the prior art described in Patent Document 1 (Japanese Patent Application Laid-Open No. 2003-84805, title of invention “Plant Load Prediction Method, Steady Plant Simulator, Optimal Plant Operation Method, and Optimal Plant Design Method”), In response to problems, plant operational costs and emissions are minimized by using metaheuristic methods such as Particle Swarm Optimization (hereinafter referred to as PSO), Genetic Algorithm (hereinafter referred to as GA), and Tabu Search (hereinafter referred to as TS). It solves an optimal operation planning problem with optimization as an objective function.
Considers discrete values such as start / stop of plant component equipment and continuous values such as fuel injection amount of component equipment, constraints on characteristics of each plant component equipment, supply and demand balance for each load type, etc. , And nonlinear optimization problems such as plant operating costs and objective functions for minimizing exhaust emissions can be solved and mathematically optimized.

JP 2003-84805 A

However, since the conventional technology described in Patent Document 1 does not directly consider the maintenance cost of each plant component device, for example, the operation cost such as fuel cost can be reduced by repeating the operation that shortens the life of the device. Even so, it is conceivable that the maintenance cost will increase and the cost will increase.
Further, in the solution search using the PSO, GA, and TS used in the prior art described in Patent Document 1, the number of plant components increases, and the search efficiency decreases as the problem order to be solved increases. There is.

  Accordingly, the present invention has been made to solve the above-described problems, and its purpose is to comprehensively consider plant operation costs, gas emissions, and maintenance costs while comprehensively considering an operation plan. An object of the present invention is to provide a plant optimum operation planning apparatus that can be efficiently created.

  A plant optimum operation planning device that solves the objective function optimum operation planning problem including minimizing plant operation costs and exhaust gas, and minimizing maintenance costs of plant components. More preferably, it is a plant optimum operation planning apparatus that uses Evolutionary Particle Swarm Optimization (hereinafter referred to as EPSO), which has better search efficiency than conventional metaheuristic techniques.

The plant optimum operation planning apparatus according to claim 1 of the present invention is:
A load predicting means for predicting various loads such as a power load, a thermal load, and an air load of the plant in an operation plan period, and generating a predicted load value;
While satisfying the supply and demand balance for each load type with respect to the predicted load value output from the load prediction means, and taking into account the restrictions on the characteristics of each plant component equipment, minimizing plant operating costs and minimizing gas emissions The objective function is to minimize the maintenance cost of each plant component device, and the start / stop state of each plant component device for each control time that is a discrete value and the fuel of each plant component device for each control time that is a continuous value Formulate the plant optimization problem as a nonlinear mixed integer programming problem with the injection amount as a state variable, and output various load prediction values for each control time, the start / stop state of each plant component device, and the fuel injection amount Optimization means;
Steady plant simulator means for sequentially calculating and outputting the start / stop state of each plant component device and the steady input / output state amount of each plant component device using the fuel injection amount;
With
The optimization means generates the start / stop state and fuel injection amount of each plant component device using the initial value of the input / output state of each plant component device, and outputs it to the steady plant simulator means,
The steady plant simulator means inputs the start / stop state of each plant component device and the fuel injection amount in the search process, and outputs the input / output state of each plant component device to the optimization means for calculation of the objective function. ,
The optimization unit generates the start / stop state of each plant component device and the fuel injection amount using the input / output state of each plant component device from the steady plant simulator unit, and outputs to the steady plant simulator unit,
Thereafter, the stationary plant simulator means and the optimization means are alternately exchanged so that the optimization means sequentially corrects the start / stop state and fuel injection amount of each plant component device, and finally close to the global optimum solution. The start / stop state and fuel injection amount of each plant component device are determined.

Moreover, the plant optimum operation planning apparatus according to claim 2 of the present invention is:
In the plant optimum operation planning device according to claim 1,
The optimization means is characterized by performing optimization by sequentially correcting the start / stop state and fuel injection amount of each plant component device using Evolutionary Particle Swarm Optimization (EPSO).

  According to the present invention as described above, a plant optimum operation plan that efficiently creates an operation plan while comprehensively considering plant operation cost minimization, gas emission minimization, and maintenance cost minimization. An apparatus can be provided.

DESCRIPTION OF EMBODIMENTS Hereinafter, a plant optimum operation planning apparatus according to the best mode for carrying out the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram of a plant optimum operation planning apparatus according to this embodiment.
The plant optimum operation planning apparatus 100 includes a load prediction unit 10, an optimization unit 20, and a steady plant simulator unit 30. Note that all of the load prediction means 10, the optimization means 20, and the steady plant simulator means 30 in FIG. 1 are realized by computer hardware and software.

  In the plant optimum operation planning apparatus 100, the load predicting unit 10 predicts various loads such as a power load, a thermal load, and an air load of the plant and generates predicted load values. The optimization unit 20 controls each control as a plan. The start / stop state of each plant component device and the amount of fuel injection for each time are passed to the steady plant simulator means 30, and the steady plant simulator means 30 side uses these input information to generate the input / output state of each device by simulation. Then, the optimization means 20 uses the input / output state of each plant component device to determine a predetermined objective function (minimization of plant operation cost, minimization of gas emission, and minimum maintenance cost). The device is searched for various start-up / stop states and fuel injection amounts that satisfy each of the optimization methods.

Each configuration will be described.
The load prediction unit 10 is a unit that predicts a load. In this embodiment, the load prediction unit 10 predicts various loads such as a power load, a thermal load, and an air load of a target plant.
For example, the load predicting means 10 performs prediction by inputting an input factor into a linear model using a conventional multiple regression equation or a nonlinear prediction model using a general neural network. For example, it is recommended to use a general neural network because the plant load has a non-linear input / output relationship, or preferably to use an improved neural network as disclosed in Patent Document 1 described above. Is done.

The load predicting means 10 will be described taking as an example the case of predicting the thermal load of the plant.
(1) Prediction model construction First, a neural network prediction model for predicting a thermal load having nonlinearity is constructed. The neural network prediction model requires learning, but as input factors that are input to the prediction model during learning, factors related to the heat load that is the output, such as hourly temperature, maximum temperature, Factors related to meteorological conditions such as minimum temperature, minimum humidity, weather, solar radiation, etc., distinction of day of the week, weekday / holiday (Saturday is included in either weekday / holiday), presence / absence and type of event, plant operation Use all or part of factors that roughly determine the heat load pattern, such as the operation pattern of the state and plant component equipment (air conditioning equipment, etc.). Actual values are used for these input data. Further, the actual value of the thermal load is used for the output, and the input factor and the output are given to the neural network to be learned by the learning method described above, and a prediction model is constructed.

(2) Prediction execution The prediction calculation of the heat load of the next day is performed using the prediction model constructed | assembled by said (1). At this time, forecast values for the next day are used as input data relating to weather conditions such as maximum temperature and minimum humidity as input factors. Moreover, what is necessary is just to use those plan values, when inputting the operation state of a plant, the operation pattern of a plant component apparatus, etc. A thermal load prediction value is generated by this prediction.

  The load predicting means 10 individually constructs and learns a neural network prediction model for predicting a thermal load having non-linearity for predicting various loads such as a power load and an air load of a target plant during an operation plan period. A predetermined input is given to generate a predicted power load value and a predicted air load value. The predicted power load value, predicted heat load value, and predicted air load value are output to the optimization unit 20 as a predicted load value.

The optimization unit 20 satisfies the energy supply (supply / demand balance) for each load type with respect to the load prediction values of the various plant loads obtained by the load prediction unit 10, and considers restrictions on the characteristics of each plant component device. While (with these as constraints), minimizing plant operating costs (energy costs such as fuel costs, received power charges, etc.) for a given period of time, for example, minimizing gas emissions (eg CO ₂ emissions), and , Maintenance cost (lubricating oil, maintenance replacement parts, maintenance personnel personnel cost and their standard operation time, start / stop frequency, etc.) as an objective function (penalty supply-demand balance imbalance and equipment characteristic constraint deviation) Set the quantity), the start / stop state of each plant component device for each control time which is a discrete value (discrete value of start or stop) and A plant optimization problem is formulated as a nonlinear mixed integer programming problem with the fuel injection amount (continuous value) of each plant component at each control time, which is a continuous value, as a state variable, and each plant component at each control time Is a means for outputting the start / stop state of the fuel and the fuel injection amount as a planned value.

  In order for the optimization means 20 to perform optimization, the input / output state of each plant device for each control time from the steady plant simulator means 30 is necessary, but initially the input / output state is not output. The initial value of each plant equipment input / output state for each control time is set in advance.At first, this initial value is input to the solution algorithm, and the start / stop state of each plant equipment for each control time and each A fuel injection amount for each control time is generated and output to the steady plant simulator means 30. There is no optimization at this stage.

  The stationary plant simulator means 30 is a state of the plant such as a connection state between plant component devices, linear or non-linear input / output characteristics, and an operation pattern that changes according to a time zone (an input / output characteristic changes according to a change in operation pattern). Using the database that stores the characteristics necessary for the calculation of the above, when the start / stop state of the plant component equipment and the fuel injection amount for the plant component equipment for each control time are obtained from the optimization means 20, each control time The steady input / output state for each plant component device (here, the transient input / output state is not considered, and only the steady state when the device output is stable) is calculated. Then, by sequentially calculating the output of a certain plant component device as the input of the next plant component device, the steady input / output state of the entire plant is finally simulated. The input / output state of the plant component device simulated in this way is sent to the optimization means 20 for calculation of the objective function for determining the optimum operation of the plant.

  For example, as will be described later as an embodiment, as shown in FIG. 2, GTG (gas turbine generator) 101, gas turbine 102, exhaust gas boiler 103, boiler 104, high-pressure header 105, heat storage layer 106, oil-cooled hot / cold water heater Suppose that an energy plant is assumed, which includes an air load, an electric load, a steam load, and a heat (air conditioning) load. The stationary plant simulator means 30 is provided with the configuration states, input / output characteristics, and operation patterns of the respective devices 101 to 109 as shown in the figure. When the fuel injection amount to the gas fired refrigerator 108 and the start / stop states of the devices 101 to 109 are given, for example, the amount of exhaust gas output from the gas turbine 102 and the amount of steam output from the boiler 104 are calculated. To do. Then, using these as inputs to the exhaust gas boiler 103, the high-pressure header 105, and the like in the next stage, the output of each device is sequentially calculated to simulate the steady input / output state of the entire energy plant.

  In such a steady plant simulator means 30, linear or non-linear input / output characteristics of each plant constituent device as shown in FIG. 2 are expressed by, for example, a neural network, and the like according to the input / output characteristics and the time zone. In addition, an operation pattern of each device (including characteristics such as time until a steady state is reached after startup) and the like are provided in advance as a database. The input / output of each plant component device from the fuel injection side toward the end of the plant is input using the fuel injection amount for each control time output from the optimization means 20 and the start / stop state of each device as input information. The steady input / output state of the entire plant is simulated by sequentially calculating the state (the amount of input / output exhaust gas, steam, water, etc. and their input / output times). This input / output state is output to the optimization means 20.

  The optimization unit 20 obtains the input / output state of each plant component device as an output from the steady plant simulator unit 30. From here, the optimization means 20 corrects the planned values of the start / stop state and the fuel injection amount by the optimization method, and outputs them to the steady plant simulator means 30.

In the following, formulation of optimization in the optimization means 20 is shown.
1. State variables State variables are the following plant quantities.
(1) Start / stop state of each plant component device (for example, gas turbine, boiler, refrigerator, cogeneration, etc.) for each control time (2) Fuel injection amount for each control time (fuel injection amount to each device above) )

2. Objective function The objective function is assumed to consist of the following terms.
(1) Minimization of plant operating costs (assuming f ₁ )
The plant operation cost is calculated as an energy cost such as a fuel charge for the plant component equipment and a received power charge.
(2) Minimizing CO ₂ emissions (assuming f ₂ )
(3) minimizing the maintenance costs of the plant constituent (a f ₃₎
The maintenance cost of the plant component equipment is calculated for each plant component equipment based on the lubricating oil cost per unit amount with respect to the operation time or the number of start / stops, the maintenance replacement parts cost, the maintenance personnel cost, etc. However, since it may be difficult to set these cost unit prices depending on the equipment, in such a case, the criteria itself such as the operation time or the number of start / stops are calculated as equivalent maintenance costs. At this time, the dimension is adjusted by applying a weighting coefficient to each quantity.
(4) penalty (supply balance imbalance, device characteristics constraint violations amount) (a _{f 4)}
In practice, the following objective function consisting of the above terms is used.

[Equation 1]
f = w ₁ f ₁ + w ₂ f ₂ + w ₃ f ₃ + w ₄ f ₄

Here, w _i is a weight for each term, and is arbitrarily set.

3. Constraint (1) Supply / demand balance for each load type This is a constraint on the balance between demand and supply for each load type such as power system, thermal system, air system, etc., and output from the load prediction means 10 when determining this supply / demand balance The estimated load value is taken into account.
(2) Restrictions on characteristics of each equipment These are restrictions on characteristics such as input / output limits and start / stop times of each plant component equipment.

4). Solving algorithm As an optimization method to search for the optimum solution of the start / stop state and the fuel injection amount while using the calculation result (input / output state of each plant component device for each control time) by the steady plant simulator means 30 Uses a metaheuristic optimization technique. For example, PSO (Particle Swarm Optimization) or EPSO (Evolutionary Particle Swarm Optimization) can be used. This algorithm for finding solutions will be described in detail later.

The optimization means 20 generates the steady state by generating the start / stop state of the plant component equipment and the fuel injection amount optimized by the solving algorithm using the input / output state of each plant constituent equipment from the steady plant simulator means 30. Output to the plant simulator means 30.
Thereafter, the stationary plant simulator means 30 and the optimization means 20 are alternately exchanged so that the optimization means 20 sequentially corrects the start / stop state and fuel injection amount of each plant component device, and finally close to the global optimum solution. The optimum operation method of the plant is determined by determining the start / stop state of each plant component and the fuel injection amount.
The plant optimum operation planning apparatus 100 is as described above.

  In such a plant optimum operation planning apparatus 100, maintenance cost minimization is particularly added as an objective function. Therefore, an increase in maintenance cost is suppressed, which contributes to overall cost reduction.

Subsequently, PSO and EPSO used for the solution finding algorithm will be described.
First, PSO will be described.
PSO is one of the metaheuristic methods developed through the simulation of a simplified social model, and was developed through expressing the movement of a flock of birds in a two-dimensional space of continuous variables. In PSO, the position (state quantity) of each agent (the above-mentioned bird) is expressed by x and y coordinates, and the velocity (vector) corresponding to the change in the position (state quantity) is represented by v _x (velocity in the x direction). , V _y (velocity in the y direction). From these position and velocity information, the position of each agent at the next time point can be updated. Based on this concept, the following optimization can be considered if the whole flock of birds takes action that optimizes some objective function.

That is, each agent remembers the individual best value (pbest) of the objective function in each search and the x and y coordinates indicating the position (state quantity). In addition, each agent shares the best value (gbest) information of the objective function so far in the group, that is, the best one in the group among pbest. Each agent attempts to change the direction to the position where pbest and gbest exist according to the current x and y coordinates and the speeds v _x and v _y and the distance between pbest and gbest. The action to be changed is expressed by correcting the speed. Using the current speed and pbest and gbest, the speed of each agent is modified by

[Equation 2]
^{_{v i k + 1 = wv i}} + c 1 rand () (pbestt i -s i k) + c 2 rand () (gbest-s i k)

v _i : speed of agent i, rand (): uniform random number from 0 to 1, s _i ^k : position of k-th search of agent i (search point), pbest _i : pbest of agent i, w: agent's Weight function for speed, c _i : Weight coefficient for each term

  By using Equation 2 above, a speed that probabilistically approaches each agent's best solution and group's best solution is obtained, and the current position (search point) of each agent is corrected by the following equation. To do.

[Equation 3]
s _i ^{k + 1} = s _i ^{k + 1} + v _i ^{k + 1}

Here, in accordance with the present invention, each agent corresponds to a plant component device, and the position S _i of each agent in Equations 2 and 3 indicates the start / stop state (discrete value) and fuel of each plant component device. It corresponds to the injection amount (continuous value), and the velocity v _{i of} each agent corresponds to the amount of change.

  PSO is a multi-point search with a plurality of search points like genetic algorithm (GA), etc., and by changing each search point stochastically using pbest of each search point and gbest of a group, This is a method for obtaining a global optimum solution (best solution). In addition, a global search that maintains and uses the current speed (the first term on the right side of (Formula 2)) and a local search that tries to approach them using pbest and gbest (the second side on the right side of (Formula 2), respectively) This is a search method having a mechanism for performing (3)) in a well-balanced manner. Furthermore, in the PSO, it is necessary to evaluate the objective function value at each step of the search, but there is an advantage that the number of evaluations need only be the number of agents regardless of the scale of the problem.

  Next, EPSO will be described. Specifically, EPSO and its improved technique are used. Here, EPSO and its improved method are as follows: V. Miranda N. Fonseca, "ESPO-BEST-OF-TWO-WORLDS META-HEURISTIC APPLIED TO POWER SYSTEM PROBLEMS", Proc. Of the Congress on Evolutionary Computation (CEC2002), It means the EPSO method described in 2002 and its improved method.

(1) State expression In EPSO, a state is expressed by a fuel injection amount to each plant component device for each control time, and a data string representing a start / stop state of each component device whose start / stop state can be changed.
(2) Algorithm for solving problems As its name suggests, EPSO is a technique that combines evolution strategies (hereinafter referred to as ES) and PSO.
The technique that enables EPSO to be applied to nonlinear mixed integer programming problems is described by Fukuyama et al. “Study of Voltage Reactive Power Control Method by Particle Swarm Optimization Considering Voltage Reliability”, IEEJ Transactions B, Vol. 119, No. 12 (1999). (December) etc., and the method for PSO is used.

Next, EPSO is a technique that combines evolution strategy ES and PSO as mentioned before, and this idea is based on giving clear selection procedure and parameter self-adaptation to PSO. . V. Miranda, who proposed this method, classifies EPSO as a special type of self-adaptive (μ, λ) -ES.
An evolution strategy is an optimization method that mimics the evolution of a living organism, similar to a genetic algorithm. Proposed by Rechenberg, Germany in the 1960s, fundamental improvements were made by Schwefel. Mainly for optimization of n-dimensional continuous functions. In (1 + 1) -ES, which is the simplest ES, there is one individual (search point), and one that is superior from a total of two individuals, a child individual generated by mutation and an original individual (parent individual) Select. Mutation becomes the main search operator, and the algorithm flow is as follows.

(1) Initialization (2) Evaluation
Repeat until the termination condition is satisfied (3) Mutation
(4) Evaluation
(5) Selection (or selection)

  The method of using a plurality of individuals (search points), that is, generating λ child individuals from μ parent individuals and determining the next individual (search point) is mainly (μ + λ) −ES and (μ, λ ) -ES. (Μ + λ) -ES is a method for selecting an excellent individual from both parent and child as the next search point, and (μ, λ) -ES is a method for selecting the next search point from only the child individual. . In (μ + λ) -ES, the search may stagnate depending on the parameters, but in (μ, λ) -ES, the life span of the individual is short, so that inappropriate parameters can be forgotten. Therefore, an effect of escaping from the local solution can be expected, and the performance is superior to (μ + λ) −ES.

An individual that is an ES search point is accompanied by a vector of design variables representing a solution and a vector of standard deviation σ of a normal distribution. This standard deviation defines the size of the mutation. Mutations follow a normal distribution and are applied for each design variable and standard deviation. Parameters such as standard deviation are called strategic parameters, and design variables are called object parameters.
Self-adaptive ES refers to a method of autonomously adjusting strategy parameters during the search process.

The outline procedure in EPSO is shown below.
(1) Replication: Each agent is replicated R times.
(2) Mutation: Each agent has a mutated weighting factor.
(3) Breeding (REPRODUCTION): A mutated agent creates offspring using a PSO search formula (moves through the solution space)
(4) Evaluation (EVALUATION): Each descendant evaluates its objective function value.
(5) SELECTION: A probable tournament leaves the best-rated agent for the next iteration.
The search formula for EPSO is shown below.

[Equation 4]
s _i ^{k + 1} = s _i ^k + v _i ^{k + 1}

[Equation 5]
^{_{v i k + 1 = w i0}} * v i k + w i1 * (pbest i -s i k) + w i2 * (gbest * -s i k)

  These look almost the same as the original PSO. However, the weighting factors are subject to the following mutations:

[Equation 6]
w _ik ^* = w _ik + τN (0,1)

Here, N (0,1) is a random number according to a Gaussian distribution with an average of 0 and a variance of 1.
Gbest is also subject to random sway as follows.

[Equation 7]
gbest ^* = gbest + τ'N (0, 1)

  Here, τ and τ ′ are learning parameters, which are fixed or treated as strategy parameters.

The EPSO algorithm is shown below.
Step.1 Determine the initial value of each agent (randomly).
Step.2 Create R copies for each agent, and mutate their weighting factors using Equation 6.
Step.3 Search points are updated with the search formula (Formula 4 and Formula 5) for the created replication agent.
Step.4 Evaluate the objective function value of each agent and duplicated agent.
Step.5 Select agents to be left for the next search by probabilistic tournament with each agent and duplicated agents.
Step.6 Update pbest and gbest. That is, the best objective function value f (pbestmin) among all pbests of agents is compared with f (gbest) so far, and if best (fbestmin)> f (gbest), gbest is updated.
Step.7 Update the search point of each agent by the search formula (Formula 4 and Formula 5).
Step.8 If the number of iterations has reached the set maximum number of iterations, it will end. If not, repeat Steps 2-7.

The overall algorithm of the optimum operation method of the plant when EPSO is used is shown below.
Step.11 Input of data (1) Input of predicted load value ・ Input predicted load value for each control time for each load type. This predicted load value is a predicted value of power load, heat load, air load, etc. of the plant.
(2) Input of plant information (these are stored in advance as a database)
・ Input / output characteristics (linear, non-linear) for each plant component and operation pattern for each control time ・ Connection relationship between each component and its characteristic equation ・ Fuel cost for each nuclear fuel type Enter as.
(3) Input of information on EPSO / Input the number of agents, each optimization parameter value, and the maximum number of searches.

Step.12 Initial value generation (1) The following values for each control time are randomly calculated for each agent.
・ Start / stop status (discrete values) for each plant component
-Of the above plant components, the fuel injection amount of fuel injection devices (continuous amount)
(2) Calculation of plant state for each agent and calculation of evaluation value. Using the start / stop state and fuel injection amount for each plant component calculated in (1), each control is performed by the steady plant simulator means 30. Obtain the input / output status of each plant component for each hour.
Further, an evaluation value is calculated for each agent using the objective function and the constraint conditions shown in FIG. Here, the evaluation value is the sum of plant operation costs, CO ₂ emissions, and maintenance costs.
(3) Initial setting of pbest and gbest • The evaluation value for each agent calculated in (2) above is used as the current pbest value for each agent.
The best value (minimum value) of the above pbest is set as gbest. As a result, initial values of pbest and gbest are set.

Step.13 Correction of the position (search point) of each agent (1) R copies are created from each agent, and their weight coefficients are mutated by Equation 6.
(2) For each agent and the created replication agent, the search point is updated by a search formula (Formula 4 and Formula 5).

Step.14 Evaluation of each agent (1) Calculation of plant state and evaluation value for each agent ・ Stationary plant using start / stop state and fuel injection amount for each plant component calculated in Step 13 The simulator means 30 obtains the input / output state of each plant component device for each control time.
-An evaluation value is calculated for each agent using the objective function and the constraint conditions shown in FIG.
(2) Selection of agents to be left for the next search-Each agent and the agent created in step 13 are used to select an agent to be left for the next search by a probability tournament.
(3) Correction of pbest and gbest. If the evaluation value for each agent calculated in (1) is better than the current pbest value for each agent, the evaluation value is set as pbest for each agent.
If the best value of the above pbest is better than the current gbest value, the best value is set as gbest. Through these processes, the optimum solution of the start / stop state of each plant component and the fuel injection amount is obtained.

Step.15 End condition check (1) End when the number of searches reaches the maximum number of searches. If not, return to Step 13.

An example in which the present invention is applied to the plant shown in FIG. 2 is shown below.
The target plant has three types of loads: an electric power load, a steam load, and a heat load. The plant includes a GTG (gas turbine generator) 101, a gas turbine 102, an exhaust gas boiler 103, a boiler 104, a high-pressure header 105, a heat storage layer 106, an oil fired hot and cold water machine 107, a gas cold and hot water machine 108, and a steam absorption refrigerator 109. , An energy plant composed of

The state variables are the start / stop states of the oil-cooled water heater / heater 106, the gas water cooler / heater 107, and the vapor absorption refrigerator 108, and the operation states of the gas turbine 102, the exhaust gas boiler 103, and the boiler 104 are determined by automatic control. It is treated as a constraint condition. The objective functions are plant operation cost minimization, CO ₂ emission minimization, and maintenance cost minimization. In this example, the number of start / stop times of the target device is set as the maintenance cost.
Table 1 shows a comparison between the case where the minimization of the maintenance cost (number of times of starting and stopping) is not considered and the case where the objective function is considered.

  Thereby, it turns out that the cost reduction can be realized in the optimum plant operation planning apparatus 100 of the present invention by considering the maintenance cost minimization.

  Subsequently, Table 2 shows a cost comparison between the case where PSO is used as a solution algorithm and the case of the present invention (using EPSO).

  Thereby, in the plant optimal operation planning apparatus 100 of this invention, when EPSO is employ | adopted especially in the solution algorithm of the optimization means 20, it turns out that there is a 3.3% cost reduction effect.

According to the present invention, it is possible to create a plant optimum operation plan that comprehensively considers plant operation costs, gas emissions, and plant maintenance costs, which contributes to the reduction of the overall costs and environmental burden associated with the plant. it can.
Further, according to the present invention, it is possible to create an operation plan that further reduces the cost, particularly when EPSO is used.

It is a block diagram of the plant optimal operation planning apparatus of the best form for implementing this invention. It is explanatory drawing of the example of an energy plant.

Explanation of symbols

100: Plant optimum operation planning device 10: Load prediction means 20: Optimization means 30: Steady plant simulator means

101: GTG (gas turbine generator)
102: Gas turbine 103: Exhaust gas boiler 104: Boiler 105: High-pressure header 106: Heat storage layer 107: Oil tank cooling / heating machine 108: Gas cooling / heating machine 109: Steam absorption refrigerator

Claims (2)

A load predicting means for predicting various loads such as a power load, a thermal load, and an air load of the plant in an operation plan period, and generating a predicted load value;
While satisfying the supply and demand balance for each load type with respect to the predicted load value output from the load prediction means, and taking into account the restrictions on the characteristics of each plant component equipment, minimizing plant operating costs and minimizing gas emissions The objective function is to minimize the maintenance cost of each plant component device, and the start / stop state of each plant component device for each control time that is a discrete value and the fuel of each plant component device for each control time that is a continuous value Formulate the plant optimization problem as a nonlinear mixed integer programming problem with the injection amount as a state variable, and output various load prediction values for each control time, the start / stop state of each plant component device, and the fuel injection amount Optimization means;
Steady plant simulator means for sequentially calculating and outputting the start / stop state of each plant component device and the steady input / output state amount of each plant component device using the fuel injection amount;
With
The optimization means generates the start / stop state and fuel injection amount of each plant component device using the initial value of the input / output state of each plant component device, and outputs it to the steady plant simulator means,
The steady plant simulator means inputs the start / stop state of each plant component device and the fuel injection amount in the search process, and outputs the input / output state of each plant component device to the optimization means for calculation of the objective function. ,
The optimization unit generates the start / stop state of each plant component device and the fuel injection amount using the input / output state of each plant component device from the steady plant simulator unit, and outputs to the steady plant simulator unit,
Thereafter, the stationary plant simulator means and the optimization means are alternately exchanged so that the optimization means sequentially corrects the start / stop state and fuel injection amount of each plant component device, and finally close to the global optimum solution. A plant optimum operation planning apparatus for determining a start / stop state of each plant component and a fuel injection amount. In the plant optimum operation planning device according to claim 1,
The plant optimization operation planning apparatus characterized in that the optimization means optimizes by sequentially correcting the start / stop state and fuel injection amount of each plant component device using Evolutionary Particle Swarm Optimization (EPSO).

Images