JP4338610B2 - Plant optimum operation plan creation system - Google Patents

Description

  The present invention is an energy plant that supplies cold and hot heat and electric power to a load, and an optimal plant operation plan that uses an optimization method to create an operation plan that minimizes plant operation costs while taking into account prediction errors such as heat load and electric power load. Regarding the creation system.

  Conventionally, various methods for obtaining optimum operation of a plant by using an optimization method in consideration of starting and stopping of various plant constituent devices have been proposed. In these conventional technologies, the input / output characteristics and operation patterns of plant components are handled as linear general equations and formulated as a mixed integer programming problem, or the problem is solved, or the input / output characteristics of plant components are In general, a method that determines the optimal operation of a plant by heuristic methods that can handle nonlinear problems, such as expert systems and fuzzy inference, taking into account non-linearity (including discontinuous and non-differentiable) and operation patterns. It was.

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. is there. Therefore, the optimal operation of the plant can be formulated as a nonlinear mixed integer programming problem that is the most difficult to solve among the optimization problems.
However, after comprehensive consideration of discrete values such as start / stop of plant component equipment and continuous values such as fuel injection amount of component equipment, restrictions on the 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 plant operating costs and exhaust gas minimization, it is difficult to mathematically optimize for problems with non-linearities.

  For this reason, in Patent Document 1 below, modern heuristic methods such as Particle Swarm Optimization (hereinafter referred to as PSO), genetic algorithm (hereinafter referred to as GA), tab search (hereinafter referred to as TS), and the like are dealt with. Is used to create an optimal operation plan that minimizes plant operation costs.

JP 2003-84805 A (Claims 3 and 4, paragraphs [0085] to [0119], FIG. 12, etc.)

By using the prior art described in the above-mentioned Patent Document 1, various plant component devices are considered after comprehensively considering discrete values such as start / stop of plant component devices and continuous values such as fuel injection amount of component devices. By mathematically optimizing problems with non-linearity using constraints on the characteristics of the system, constraints such as supply and demand balance for each load type, and objective functions such as plant operation costs and exhaust gas minimization The optimal operation plan can be created.
In this prior art, an optimum operation plan is created based on predicted load values such as power load and heat load predicted using a neural network, for example.

However, since there is always a prediction error in the load prediction value, when this prediction error is considered, the operation plan created by the conventional technique of Patent Document 1 is not necessarily the optimum operation plan that minimizes the operation cost. Is not limited.
Therefore, the problem to be solved by the present invention is to provide a plant optimum operation plan creation system that can create an operation plan that minimizes operation costs in consideration of prediction errors of plant loads such as power load and heat load by an optimization method. There is.

In order to solve the above problem, the invention described in claim 1
As a plan, when the start / stop state of each plant component device for each control time and the fuel injection amount are input, the input information, the predicted load value for each control time, and the constraints on the characteristics of each device Using a plant simulator unit that creates and outputs a first operation plan assuming that there is no prediction error, including the start / stop state and fuel injection amount of each device for each control time,
Predetermining the load prediction error at the representative point of the prediction error assuming that there is a prediction error,
A local control unit that gives the load prediction error to the first operation plan, corrects the operation plan by local control of each device, and outputs the corrected operation plan as a second operation plan based on the prediction error together with the first operation plan When,
The occurrence probability of the representative point of the prediction error when it is assumed that there is no prediction error, and the occurrence probability of the representative point of the prediction error when it is assumed that there is a prediction error, obtained in advance from the probability distribution of the load prediction error. As the load state occurrence probability in each operation plan, the operation cost expected value is calculated using the operation cost in the first operation plan, the operation cost in the second operation plan, and the load state occurrence probability in each operation plan. A cost calculator to
The operation cost expectation value is set as the objective function of the optimization problem, and the start / stop state and the fuel injection amount of each device for each control time that minimizes the objective function are searched by the metaheuristic optimization method, and these And an optimization unit that outputs a start / stop state and a fuel injection amount to the plant simulator unit as a plan .

The invention described in claim 2 is the plant optimum operation plan creation system according to claim 1,
Assuming that the probability distribution of load prediction error at each representative point is a normal distribution,
Obtain the standard deviation of the load prediction error from the past load prediction value and the actual value,
Determining each representative point and its load prediction error based on the standard deviation;
The occurrence probability of each representative point is determined as an occurrence probability determined by the representative point and the normal distribution .

  According to the present invention, it is possible to create a plant optimum operation plan in consideration of a load prediction error, and it is possible to contribute to a reduction in plant operation cost. In addition, there is an effect that an operator engaged in actual operation can easily compare and recognize the difference in operation plan between when the load prediction error is considered and when it is not considered.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a conceptual configuration diagram of a plant optimum operation plan creation system considering a load prediction error according to the present embodiment. This system includes an optimization unit 10, a plant simulator unit 20, a local control unit 30, a cost. The calculation unit 40 is configured. Each of these components includes computer system hardware and software.

The optimization unit 10 outputs to the plant simulator unit 20 the start / stop state of each plant component device and the fuel injection amount for each control time, which are state variables, as a plan.
The plant simulator unit 20 uses these input information and load prediction values (load prediction values not considering prediction errors) for each control time generated separately, and further, input / output characteristics and operation modes of each device. Etc. is used to create a plant operation plan (operation plan without consideration of prediction error: first operation plan) that includes the start / stop state of each device for each control time and the amount of fuel injection. Output to 30.

Next, the local control unit 30 gives a load prediction error to the operation plan without error consideration (first operation plan) created by the plant simulator unit 20, and corrects the operation plan by local control of each device, The corrected operation plan is output together with the first operation plan as an operation plan based on the load prediction error (second operation plan).
The local control will be described later.

Here, in an actual plant, the arrangement of pipes is determined based on a previously examined configuration. FIG. 2 shows an example of the piping arrangement of the plant.
That is, the heat output generated together with the electric power by the generator 101 passes through the pipe 102 and is a GENELINK (exhaust heat input type gas absorption chiller / heater) 103 as a heat supply device, a heat exchanger 104 for heating, and heat exchange for hot water supply. The heat is supplied in series in the order of the vessel 105, and the excess heat is processed by the cooling tower 106. At this time, the amount of heat supplied to each heat supply device is controlled for each operation plan period by the flow rate control valves 107, 108, 109 provided in parallel to each device.

Here, when the magnitude of the thermal load is different between the predicted value and the actual value (that is, when there is a prediction error), the flow rate control valves 107, 108, 109 are controlled online to each heat supply device. It is difficult to balance the supply and demand of each load by changing the amount of heat supplied. In view of this, in the plant component equipment, when there is a difference between the predicted load value and the actual value, control performed by the equipment alone to balance the supply and demand balance is determined in advance. Control that balances the supply and demand balance with this device alone is defined here as local control.
The local control unit 30 shown in FIG. 1 simulates local control performed by the plant component equipment itself.

As described above, the local control unit 30 assumes that the load has a prediction error, corrects the operation plan without considering the prediction error (first operation plan) as a part of the local control, and considers the load prediction error. The second operation plan (second operation plan) is created.
Note that the method of giving the load prediction error to the local control unit 30 assumes that the distribution of the load prediction error is a normal distribution, and gives the error representative points as three points of error zero, + error, and -error.

FIG. 3 shows a load prediction error assumed to be a normal distribution. The errors are given as 0 (error zero), + 0.43σ (+ error), and −0.43σ (−error). Here, σ is a standard deviation, and is a value that can be easily obtained from the past load predicted value and the actual value.
+ 0.43σ assumed as a representative point of + error has a probability of 0.3333 or higher when the error is + 0.43σ or more, and −0.43σ set as a representative point of −error has an error of −0. This means that the probability of being 43σ or less is 0.3333.

Therefore, the probability of becoming zero, which is a representative value of zero error, is 0.3333, and it is determined in consideration that all three representative points have a probability of 0.3333. These error representative points and their occurrence probabilities can be arbitrarily determined.
In this way, in this embodiment, the values of the three load prediction error representative points and the probability that the prediction error will occur are arbitrarily given.

As described above, in the present embodiment, the plant simulator unit 20 creates an operation plan without error consideration (corresponding to zero error in FIG. 3) and outputs the operation plan to the local control unit 30. The remaining two states, that is, an operation plan when there is a + error and an operation plan when there is a -error are created as operation plans based on the prediction error.
Therefore, a total of three operation plans (one first operation plan and two second operation plans) are created and output by the local control unit 30.

  Next, the cost calculation unit 40 calculates an operation cost for each operation plan created by the local control unit 30. That is, the cost calculation unit 40 calculates each operation cost from the operation plan without error consideration and each operation plan when there is + error and −error, and these operation cost and prediction error. Using the load state occurrence probability according to the presence / absence of the operating condition, an expected operating cost value is obtained by an arithmetic expression described later. Then, using this expected operating cost as an objective function of the optimization problem, the start / stop state of each device and the fuel injection amount for each control time that minimizes this objective function are searched.

The optimization formulation in this embodiment is shown below.
(1) State variables The state variables are the following plant quantities.
a. Start / stop state (discrete value) of each plant component device (for example, gas turbine, boiler, refrigerator, cogeneration, etc.) for each control time
b. Fuel injection amount for each control time (continuous value)

(2) Objective function The objective function is the expected operating cost shown in the following Equation 1, which is minimized.
[Formula 1]
F = P ₀ C ₀ + P ₊ C ₊ + P ₋ C ₋
Where F: objective function (expected operating cost)
P ₀ : Load state occurrence probability when load prediction error is zero
P ₊ : Load state occurrence probability when load prediction error is positive (+ error)
P ₋ : Load state occurrence probability when load prediction error is negative (−error)
C ₀ : Operating cost when load prediction error is zero
C ₊ : Operating cost when load prediction error is positive (+ error)
C ₋ : Operating cost when load prediction error is negative (−error)

(3) Constraints a. Supply-demand balance for each load type This is a constraint condition for maintaining the balance of energy demand and supply for each load type, such as power system load and heat system load.
b. Constraints on equipment characteristics These are restrictions on the characteristics determined by the input / output limits and start / stop times of each plant component.

(4) Solution-solving algorithm As an optimization method, a meta-heuristic (modern heuristic) optimization method is used. Specifically, the aforementioned PSO and its improved technique, GA and its improved technique, TS and its improved technique, Ant Colony Optimization (abbreviated as ACO) and its improved technique, and the like are used.
Details of these algorithms are described in Patent Document 1 described above. Here, FIG. 4 shows an algorithm for finding a solution using PSO as an example.

Step S1: Reading input data Optimization calculation such as predicted load value, load prediction error, input / output characteristics (plant characteristics) of each plant equipment, fuel cost, parameters required for PSO (number of agents, maximum number of searches), etc. Enter the necessary information.
Step S2: Generation of Initial Operation State A plurality of initial states of plant operation are randomly created with a load prediction error of 0 for each agent.

· Step S3: calculating operational costs C ₀ operation state without considering the load prediction error created for each cost calculation agent operation state without considering the load prediction error (this operational state corresponds to a first operating strategy), respectively To do.
-Step S4: Correction of operation status due to + error When + error is given from the operation status that does not take into account the load prediction error created in step S2, the operation status is corrected so that the supply and demand balance is balanced according to local control. (The corrected operation state corresponds to the second operation plan), and an operation cost C ₊ is calculated.

-Step S5:-Correction of operation state due to error-When the error is given from the operation state that does not consider the load prediction error created in step S2, the operation state is corrected so that the supply and demand balance is balanced according to local control. (operation state after modification corresponds to the second operating strategy), the cost C operation _- is calculated.

Step S6: Objective function calculation The expected operating cost is calculated by the above-described equation 1 using each operating cost and the occurrence probability of each load state created in steps S4, 5, and 6.
Step S7: Save pbest (personal best) For each agent, the state having the highest evaluation value in the search so far is saved as pbest.
Step S8: Save gbest (group best) The state having the highest evaluation value so far among all agents is saved as gbest.

Step S9: Checking the end condition When the number of searches reaches the maximum number of searches, the process ends. Otherwise, the operation state is changed by changing the state of each agent to move the search point, and the process returns to step S3 (step S11).
Step S10: Output of the solution The operation state of gbest up to the present is output as an optimal solution (optimal operation plan).

This embodiment was applied to the plant shown in FIG.
This target plant has three types of loads: electric load, cooling load, and heating load. The electric load includes electric power purchased from an electric power company and a specific scale electric power company (Er = Et + Ep) and CGS (cogeneration system). ) supplied by a generator power _E g1 ₊ ...... + _{E gNg} by the generator _G 1 _{~G Ng.} Here, Ep is the power supplied from a specific scale electric power company (PPS).
The cooling load is supplied from the GENELINK (exhaust heat input type gas absorption chiller / heater) GL, and the heating load is supplied from the heating heat exchanger HEXh and the heating boiler Bh. Further, if the heat from the generator _G 1 _{~G Ng} is left over, it is adapted to be waste heat through the radiator _CT 1 _{to CT Ng.}

The equipment configuration of the target plant is as follows.
・ Generator: 750 kW x 2 units ・ GL: 4700 kW x 1 unit ・ Heating heat exchanger: 1 unit ・ Heating boiler: 1 unit ・ Heat radiator: 1050 kW x 2 units Here, the load is an error every control time Zero, + error, and -error were set as shown in Table 1.
Further, the occurrence probability of the error zero, + error, and -error states (load state occurrence probability) was 1/3, respectively.

  Table 2 shows the expected operating cost value for each load for each control time without considering the load prediction error, and the expected operating cost value for each load for each control time considering the load prediction error as + error and -error. Shown in comparison. In Table 2, the expected operating cost when the load prediction error is considered is based on a relative comparison with the expected operating cost when the load prediction error is not taken as 100. This expected operating cost value is calculated by the cost calculation unit 40 of FIG.

  From Table 2, it can be seen that when the load prediction error is taken into consideration, there is a cost reduction effect of 0.03% as compared with the case where the load prediction error is not taken into consideration.

  FIG. 6 shows a cooling load operation plan, FIG. 7 shows a heating load operation plan, and FIG. 8 shows a waste heat operation plan. (A) in each figure is when the load prediction error is not taken into account (for the sake of convenience, no uncertainty is considered), and (b) is when the load prediction error is taken into account (also, the uncertainty is taken into account). It is. Note that the predicted cooling load values in FIGS. 6A and 6B are the same, and the predicted heating load values in FIGS. 7A and 7B are also the same.

In FIG. 6A in which the cooling load prediction error is not taken into consideration, for example, when paying attention to the operation plan from 8:00 to 18:00, the cooling load is entirely covered by the auxiliary heat amount of the Genelink GL from 8:00 to 16:00. At 18:00, an operation plan is created that is shared by the auxiliary heat amount and exhaust heat of the GENELINK GL. On the other hand, according to FIG.6 (b) which considered the prediction error, the operation plan which shares a part of cooling load by the exhaust heat of GENELINK GL between 8 o'clock and 15 o'clock is created. .
Further, in FIG. 7A in which the heating load prediction error is not taken into account and in FIG. 7B in which the prediction error is taken into account, for example, in the operation plan from 8:00 to 15:00, in FIG. Although the share of the heat output of the boiler (BH in the legend and corresponding to Bh in FIG. 5) is smaller than that of the heat exchanger for heating (HH in the legend and equivalent to HEXh in FIG. 5), In FIG.7 (b), the operation plan with a larger share ratio by the heat output of the heating boiler than FIG.7 (a) is created.
Further, in FIGS. 8 (a) and 8 (b) regarding the exhaust heat operation plan, from 8 o'clock in FIG. 8 (b), depending on whether or not the exhaust heat is used in FIGS. 6 (a) and 6 (b). The exhaust heat recovery amount of Genelink GL appears at 15:00.

  As described above, in this embodiment, it becomes clear that a difference occurs in the operation plan depending on whether or not an uncertain factor such as a load prediction error is taken into consideration, and it becomes easy for the operator to compare the profit and loss of both plans.

It is a notional block diagram which shows embodiment of this invention. It is a figure which shows an example of piping arrangement | positioning of a plant. It is explanatory drawing of the generation | occurrence | production probability of a load prediction error. It is a flowchart which shows the solution algorithm at the time of using PSO in embodiment. It is a block diagram of the plant to which an Example is applied. It is a figure which shows the operation plan of the cooling load in an Example. It is a figure which shows the operation plan of the heating load in an Example. It is a figure which shows the operation plan of the waste heat in an Example.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10: Optimization part 20: Plant simulator part 30: Local control part 40: Cost calculation part 101: Generator 102: Piping 103: Genelink 104: Heat exchanger for heating 105: Heat exchanger for hot water supply 106: Cooling tower 107 , 108, 109: Flow control valve

Claims (2)

As a plan, when the start / stop state of each plant component device for each control time and the fuel injection amount are input, the input information, the predicted load value for each control time, and the constraints on the characteristics of each device Using a plant simulator unit that creates and outputs a first operation plan assuming that there is no prediction error, including the start / stop state and fuel injection amount of each device for each control time,
Predetermining the load prediction error at the representative point of the prediction error assuming that there is a prediction error,
A local control unit that gives the load prediction error to the first operation plan, corrects the operation plan by local control of each device, and outputs the corrected operation plan as a second operation plan based on the prediction error together with the first operation plan When,
The occurrence probability of the representative point of the prediction error when it is assumed that there is no prediction error, and the occurrence probability of the representative point of the prediction error when it is assumed that there is a prediction error, obtained in advance from the probability distribution of the load prediction error. As the load state occurrence probability in each operation plan, the operation cost expected value is calculated using the operation cost in the first operation plan, the operation cost in the second operation plan, and the load state occurrence probability in each operation plan. A cost calculator to
The operation cost expectation value is set as the objective function of the optimization problem, and the start / stop state and the fuel injection amount of each device for each control time that minimizes the objective function are searched by the metaheuristic optimization method, and these An optimization unit that outputs a start / stop state and a fuel injection amount to the plant simulator unit as a plan,
Plant optimal operation planning system, characterized in that with. In the plant optimum operation plan creation system according to claim 1,
Assuming that the probability distribution of load prediction error at each representative point is a normal distribution,
Obtain the standard deviation of the load prediction error from the past load prediction value and the actual value,
Determining each representative point and its load prediction error based on the standard deviation;
The plant optimum operation plan creation system , wherein the occurrence probability of each representative point is determined as an occurrence probability determined by the representative point and the normal distribution .

Images