CN102980272B - Air conditioner system energy saving optimization method based on load prediction - Google Patents

Abstract

The invention discloses an air conditioner system energy saving optimization method based on load prediction. Historical data is utilized to predict the load of a future-moment air conditioner system; then, an energy consumption model of an air conditioner system is optimized by a particle algorithm to obtain an optimal operation parameter under the load condition; according to the real-time retardation time of the system, each parameter variable of the system is controlled in advance; a control time difference brought by the big time delay and big inertia of the air conditioner system is avoided; the amount of coldness supplied by the system and the amount of coldness used by the load are guaranteed to be same; and synchronism in time is synchronous realized. In addition, according to the air conditioner system energy saving optimization method based on load prediction, the load prediction and the optimization control of the air conditioner system can be carried out in real time, and therefore the air conditioner system is always under the optimal working state or the near-optimal working state to achieve the purpose of energy saving optimization.

Description

A kind of air conditioner system energy saving optimization method based on load prediction

Technical field

The present invention relates to a kind of air conditioner system energy saving optimization method, belong to the energy saving optimizing field of air-conditioning system.

Background technology

Along with the development of Chinese Urbanization and process of industrialization, construction industry development is swift and violent, and building energy consumption constantly increases, and building energy consumption has occupied the more than 27% of social total energy consumption, and some area has approached 40%, and wherein 2/3rds energy consumption is consumed by air-conditioning system.At building energy consumption, account under the ever-increasing present situation of ratio of whole energy resource consumption, the air conditioner system energy saving in building has become emphasis and the focus in energy-saving field.According to the energy-conservation concept of terminal, the dynamics that strengthens air conditioner energy saving has huge theoretical and practical significance to saving the energy.

Owing to lacking advanced control technology means and equipment, central air conditioner system still adopts traditional labor management mode and easy switching control device mostly, can not realize air-condition freezing discharge and follow the variation of end load and dynamic adjustments, when operation at part load, cause energy waste very large, make China's energy for building inefficiency.In recent years, along with application and the development of building automatic (BA) system and variable-frequency control technique, people start to adopt BA or frequency converter to control air-conditioning system equipment., by BA system, the equipment such as handpiece Water Chilling Units, water pump, blower fan are carried out to a long-range on off control, realize its " use becomes more meticulous " energy-conservation; Another kind is by the collection to air-conditioning system ductwork pressure or temperature, usings pressure reduction or the temperature difference as controlled parameter, adopts PI or PID to control the operating frequency that frequency converter regulates water pump, makes pump capacity follow controlled parameter variation, thereby reaches the object of pump energy saving.But, central air conditioner system be one have time lag, time become, the complication system of non-linear and large inertia, its complexity causes central air conditioner system to be difficult to accurately Mathematical Modeling or method are described, and, the engineering parameter of PID controller depends on commissioning staff's experience to a great extent, affected by human factor very large, once adjust, just can not follow load adjusts automatically with the variation of operating mode, not only the poor system that also easily causes of energy-saving effect is shaken, and affects the stability of central air conditioner system operation and the service quality of air conditioning terminal.In article " the air conditioning water flow dynamics control technology based on load prediction ", author adopts the method based on load prediction, utilize intelligent fuzzy control technique, according to load and environment temperature, by fuzzy rule base, to controlling, carry out intelligent decision, the method has reached energy-conservation Optimal Control effect to a certain extent, but the accuracy of the decision rule obtaining by fuzzy control still needs to improve.

Due to central air conditioner system time become behavioral characteristics, traditional Energy Saving Control strategy can not be real-time in air-conditioning system running online accumulation and comprehensive relevant information, carry out the control parameter of instant correction or regulating system, more can not make air-conditioning system all the time in optimum or approach optimum duty.On the other hand, large dead time and large inertia due to air-conditioning system, the output of control system always will just can be worked after τ through lag time, yet popular constant-pressure drop and the constant difference control model of air conditioner energy saving control field belongs to " following control " at present, generally be only applicable to controlled device or process without time lag, be used in and in air-conditioning system, be difficult to obtain good control effect.

Summary of the invention

The deficiency existing for existing air conditioner system energy saving control method, the present invention proposes a kind of air conditioner system energy saving optimization method based on load prediction.Basic thought of the present invention is, analysis by comprehensive systematic parameter detection and historical data judges, the load of etching system (chilling requirement) when Inference Forecast goes out future, then optimize the energy consumption function model of air-conditioning system, obtain the optimized operation parameter under this load, again according to real-time lag time of the τ of system, in advance each parametric variable of system is controlled, with assurance system, for cold-peace load is cold, quantitatively equate, synchronous on time, make air-conditioning system all the time in optimum or approach optimum duty.

The specific implementation step of a kind of air conditioner system energy saving optimization method based on load prediction that the present invention proposes is as follows:

Step 1, collects historical data, and data is normalized;

Step 2, utilizes the historical data after normalization, by neutral net, the air-conditioning system load Q in the air-conditioning system moment to be measured is predicted;

Step 3, the energy consumption model of setting air-conditioning system is P=f (Q, T _1o, T _2o, v ₁, v ₂, Fair), the energy consumption power that wherein P is air-conditioning system, Q is air-conditioning system load, T _1ofor air-conditioning system chilled water leaving water temperature, T _2ofor air-conditioning system cooling water leaving water temperature, v ₁for the chilled water pump flow of air-conditioning system, v ₂cooling water pump flow for air-conditioning system, Fair is the air mass flow of air-conditioning system, under the prerequisite of air-conditioning system load Q that obtains the moment to be measured by described step 2, utilize particle cluster algorithm to carry out the optimization of energy consumption model, obtain the best parameter group when the energy consumption power P in the moment to be measured is got minimum of a value, i.e. T _1o, T _2o, v ₁, v ₂, the best parameter group of Fair; Wherein particle cluster algorithm comprises the following steps:

The first step, is expressed as X by particulate _i=(x _i1, x _i2, x _i3, x _i4, x _i5), wherein, x _i1corresponding chilled water leaving water temperature T _1o, x _i2corresponding cooling water leaving water temperature T _2o, x _i3corresponding chilled water pump flow v ₁, x _i4corresponding cooling water pump flow v ₂, x _i5corresponding air mass flow Fair, setting Particle Swarm scale is s, evolutionary generation is t _max, to each particulate X _i(i=1,2 ..., s) carry out the initialization of position and speed, to i arbitrarily, j (j=1,2 ..., 5), all in its range of variables, obey equally distributed generation x _ijand v _ijinitialization value, and the local desired positions P to particulate _iwith overall desired positions P _ginitialize;

Second step, speed and the position of renewal particulate, formula is expressed as follows:

v _ij(t+1)＝ω·v _ij(t)+m ₁r ₁(t)(p _ij(t)-x _ij(t))+m ₂r ₂(t)(p _gj(t)-x _ij(t))；

(i＝1,2,…,s)(j＝1,2,…,5)

x _ij(t+1)＝x _ij(t)+v _ij(t+1)；(i＝1,2,…,s)(j＝1,2,…,5)

Wherein, v _ijand x (t) _ij(t) be respectively particulate t for time speed and position, ω is inertia weight, m ₁and m ₂for aceleration pulse, r ₁and r ₂for the random number between 0 and 1, P _i(t) be the local desired positions in t generation of particulate, P _g(t) be the t overall desired positions in generation, p _ij(t) be vectorial P _i(t) the j dimension element in, p _gj(t) be vectorial P _g(t) the j dimension element in;

The 3rd step, calculates the target function value f (X of each particulate in t+1 generation _i);

The 4th step, the local desired positions P of renewal particulate _iwith overall desired positions P _g, formula table is shown:

$P_i(t+1)=\left\{\begin{array}{cc}{P}_i\left(t\right)& f\left(X_i(t+1)\right)\&GreaterEqual;f\left(P_i\left(t\right)\right)\\ X_i(t+1)& f\left(X_i(t+1)\right)<f\left(P_i\left(t\right)\right)\end{array}\right.$

$P_g(t+1)=\left\{\begin{array}{cc}{P}_g\left(t\right)& f\left(X_i(t+1)\right)\&GreaterEqual;f\left(P_g\left(t\right)\right)\\ X_i(t+1)& f\left(X_i(t+1)\right)<f\left(P_g\left(t\right)\right)\end{array}\right.$

P wherein _i(t+1) be the local desired positions in t+1 generation of particulate, P _g(t+1) be the t+1 overall desired positions in generation, X _i(t+1) be that t+1 is for particulate;

The 5th step, if evolutionary generation t < is t _max, return to second step, otherwise finish and export overall desired positions P _g;

Step 4: according to the T obtaining _1o, T _2o, v ₁, v ₂, the best parameter group of Fair, controls in advance to each parametric variable, and when guaranteeing that moment to be measured arrives, air-conditioning system is in optimum duty.

Accompanying drawing explanation

Fig. 1 is the schematic diagram that in the present invention, the air conditioner system energy saving based on load prediction is optimized.

Fig. 2 is the network model structure chart of neutral net in the present invention.

Fig. 3 utilizes particle cluster algorithm to carry out the flow chart of air conditioning energy consumption model optimization in the present invention.

The specific embodiment

Suzhou deluxe hotel take below as embodiment, and by reference to the accompanying drawings the specific embodiment of the present invention is described in detail.

With reference to accompanying drawing 1, the method for the air conditioner system energy saving optimization based on load prediction of the present invention mainly comprises following step:

Step 1: collect the historical data line number Data preprocess of going forward side by side.

According to the factor analysis that affects air conditioning energy consumption, the historical data that the present invention collects comprises: the occupancy rate of t and air-conditioning system are loaded constantly, moment t front 24,48,72,168 hours, be t-24, t-48, t-72, the t-168 outdoor mean temperature of actual measurement and air-conditioning system load constantly, and using above-mentioned data as sample data.In order to reduce the impact of unusual sample on neutral net performance, air-conditioning system load and outdoor mean temperature in sample data are carried out respectively to following normalization, make its scope between [0,1].Normalization formula is:

$y=\frac{x-x_{\min }}{x_{\max }-x_{\min }},$

Wherein, x is sample data, and y is the sample data after normalization, x _minfor the minimum of a value of x, x _maxmaximum for x.

Step 2: neuron network simulation.

Utilize neutral net to predict the following load constantly of air-conditioning system, specifically comprise as follows:

(1) determine the input variable number of neutral net.(2) determine the output variable number of neutral net.(3) determine the hidden layer element number of neutral net.(4) netinit carry out network training.(5) utilize the network training to carry out load prediction.

In the present invention, the input variable of neutral net is: the occupancy rate of moment t, t-24, t-48, t-72, t-168 actual measurement outdoor temperature and air-conditioning system load constantly; Output variable is the air-conditioning system load of t constantly, and wherein, outdoor temperature refers to outdoor mean temperature.Then, all data messages are formed to sample data, utilize BP algorithm of neural network, according to the network structure shown in Fig. 2, carry out network training.

The input variable number of network is 9, the air conditioner load when outdoor temperature when air conditioner load when outdoor temperature when air conditioner load when outdoor temperature when air conditioner load when outdoor temperature while being respectively occupancy rate, the t-24 of t constantly, t-24, t-48, t-48, t-72, t-72, t-168 and t-168.

The output variable number of network is 1, is the air conditioner load of moment t.

The hidden layer number of network is determined by following formula

$s=\sqrt{0.43\mathrm{kn}+{0.12n}^2+2.54k+0.77n+0.35}+0.5$

Wherein, the number that k is input layer, n is the neuronic number of output layer, the number that s is hidden layer neuron.In the present invention, k=9, n=1, substitution above formula can obtain s=5.7887, therefore hidden layer is elected 6 as.After network training completes, obtain predicting the neural network model of air-conditioning system load.

Step 3: the load that utilizes neural network prediction air-conditioning system.

Utilize the neural network model generating in step 2, input the input variable information in the moment to be measured, can obtain the air-conditioning system load Q in the moment to be measured.

Step 4: utilize population to be optimized air conditioning energy consumption model.

In the present invention, whole air-conditioning system is mainly comprised of handpiece Water Chilling Units, chilled water pump, cooling water pump and four main energy consumption equipments of blower fan of cooling tower.According to equipment operation logic, can obtain corresponding power consumption power is respectively: P ₁, P ₂, P ₃and P ₄.So, the total energy consumption P=P of whole air-conditioning system ₁+ P ₂+ P ₃+ P ₄, its energy consumption model is P=f (Q, T _1o, T _2o, v ₁, v ₂, Fair), wherein affect energy consumption power because have, air-conditioning system load Q, chilled water leaving water temperature T _1o, cooling water leaving water temperature T _2o, chilled water pump flow v ₁, cooling water pump flow v ₂with air mass flow Fair.

The energy saving optimizing of air-conditioning system refers to equipment when operation, and the power consumption of whole air-conditioning system is minimum, rather than wherein the power consumption of certain equipment is minimum, and therefore, the present invention is the object function as optimization the energy consumption P of whole air-conditioning system.Then, under the known prerequisite of prediction load Q, utilize population to carry out the optimization of energy consumption model, the best parameter group when obtaining energy consumption power P and getting minimum of a value,

(T _1o,T _2o,v ₁,v ₂,Fair)。

With reference to accompanying drawing 3, the process of utilizing particle cluster algorithm to be optimized air conditioning energy consumption is as follows:

(1) parameter initialization.Be mainly to the 0th generation Particle Swarm random site x _ijand speed v (0) _ij(0) carry out initial setting.In particle cluster algorithm, a particle correspondence a solution of object function.In the present invention, moment air-conditioning system load Q to be measured is obtained by neural network prediction by step 3, and Q is known, so initialization particle is X _i=(x _i1, x _i2, x _i3, x _i4, x _i5), wherein, x _i1corresponding chilled water leaving water temperature T _1o, x _i2corresponding cooling water leaving water temperature T _2o, x _i3corresponding chilled water pump flow v ₁, x _i4corresponding cooling water pump flow v ₂, x _i5corresponding air mass flow Fair.Setting Particle Swarm scale is s, and evolutionary generation is t _max, to each particulate X _i(i=1,2 ..., s) carry out the initialization of position and speed, to i arbitrarily, j (j=1,2 ..., 5), all in range of variables, obey to be uniformly distributed producing x _ijand v (0) _ij(0).In order to reduce particle, leave during evolution the possibility of search volume, by v _ijlimit within the specific limits, i.e. [v _min, v _max].If search volume is limited to [x _min, x _max] in, set v _min=0.75x _min, v _max=0.75x _max.For example, when j=1, x _i1range of variables be [T _1omin, T _1omax], can obtain v _i1scope be [0.75T _1omin, 0.75T _1omax], respectively at range of variables [T _1omin, T _1omax] and [0.75T _1omin, 0.75T _1omax] in evenly produce x _i1and v (0) _i1(0) (i=1,2 ..., s).Local desired positions P _i(0) initialize and be set to X _i(0), (i=1,2 ..., s), overall desired positions P _g(0) initialize the P that is set to adaptive value minimum _i(0) (i=1,2 ..., s), wherein adaptive value is the target function value f (X of current particulate _i).

(2) upgrade speed and the position of particulate.Formula is as follows,

v _ij(t+1)＝ω·v _ij(t)+m ₁r ₁(t)(p _ij(t)-x _ij(t))+m ₂r ₂(t)(p _gj(t)-x _ij(t))

x _ij(t+1)＝x _ij(t)+v _ij(t+1) (i＝1,2,…,s)，(j＝1,2,…,5)

Wherein, v _ijand x (t) _ij(t) be respectively particulate t for time speed and position, ω is inertia weight, m ₁and m ₂for aceleration pulse, r ₁and r ₂for the random number between 0 and 1, P _i(t) be the t of particulate i for local desired positions, P _g(t) be the t overall desired positions in generation, p _ij(t) be vectorial P _i(t) the j dimension element in, p _gj(t) be vectorial P _g(t) the j dimension element in.

(3) calculate the adaptive value of each particulate, t+1 is for particulate X _i(t+1) target function value f (X _i).

(4) the local desired positions P of new particle more _iwith overall desired positions P _g.

For each particulate, by its adaptive value and the desired positions P living through _iadaptive value compare, if better, using it as current desired positions.For minimization problem of the present invention, target function value is less, and corresponding adaptive value is better, that is:

$P_i(t+1)=\left\{\begin{array}{cc}{P}_i\left(t\right)& f\left(X_i(t+1)\right)\&GreaterEqual;f\left(P_i\left(t\right)\right)\\ X_i(t+1)& f\left(X_i(t+1)\right)<f\left(P_i\left(t\right)\right)\end{array}\right.(i=\mathrm{1,2},...,s)$

Overall desired positions P in colony _g, P _g(t) ∈ { P ₀(t), P ₁(t) ..., P _s}, and f (P (t) _g(t))=min{f (P ₀(t)), f (P ₁(t)) ..., f (P _s(t)) }.For each particulate, by its adaptive value and the overall desired positions P living through _gadaptive value compare, if better, using it as current overall desired positions.That is:

$P_g(t+1)=\left\{\begin{array}{cc}{P}_g\left(t\right)& f\left(X_i(t+1)\right)\&GreaterEqual;f\left(P_g\left(t\right)\right)\\ X_i(t+1)& f\left(X_i(t+1)\right)<f\left(P_g\left(t\right)\right)\end{array}\right.$

(5) if evolutionary generation t < is t _max, return to step (2), otherwise finish and export optimal solution P _g.

Step 5: according to optimized operation parameter, each parametric variable is controlled in advance.

The optimal solution P obtaining in step 4 _gcorresponding air-conditioning system the consume energy best value of respectively controlling parameter when minimum, i.e. optimum parameter combinations (T _1o, T _2o, v ₁, v ₂, Fair).Then according to real-time lag time of the τ of system, in advance each parametric variable of system is controlled, when guaranteeing following t constantly, system quantitatively equates for cold-peace load is cold, synchronous on time, make air-conditioning system all the time in optimum or approach optimum duty, reach the object of energy saving optimizing.

The present invention utilizes historical data, dope the following load of air-conditioning system constantly, then optimize the energy consumption model of air-conditioning system, obtain the optimized operation parameter under this loading condiction, according to real-time lag time of the τ of system, in advance each parametric variable of system is controlled again, not only avoided control time of bringing due to large dead time and the large inertia of air-conditioning system poor, and the system that guaranteed is cold quantitatively equal for cold-peace load, synchronous on the time.On the other hand, the present invention can also carry out load prediction and the optimal control of air-conditioning system in real time, makes air-conditioning system all the time in optimum or approach optimum duty, thereby reaches the object of energy saving optimizing.

Claims (2)

1. the air conditioner system energy saving optimization method based on load prediction, mainly comprises following concrete steps:

Step 1, collects historical data, and data is normalized;

Step 2, utilizes the historical data after normalization, by neutral net, the air-conditioning system load Q in the air-conditioning system moment to be measured is predicted;

Step 3, the energy consumption model of setting air-conditioning system is P=f (Q, T _1o, T _2o, v ₁, v ₂, Fair), the energy consumption power that wherein P is air-conditioning system, Q is air-conditioning system load, T _1ofor air-conditioning system chilled water leaving water temperature, T _2ofor air-conditioning system cooling water leaving water temperature, v ₁for the chilled water pump flow of air-conditioning system, v ₂cooling water pump flow for air-conditioning system, Fair is the air mass flow of air-conditioning system, under the prerequisite of air-conditioning system load Q that obtains the moment to be measured by described step 2, utilize particle cluster algorithm to carry out the optimization of energy consumption model, obtain the best parameter group when the energy consumption power P in the moment to be measured is got minimum of a value, i.e. T _1o, T _2o, v ₁, v ₂, the best parameter group of Fair; Wherein particle cluster algorithm comprises the following steps:

The first step, is expressed as X by particulate _i=(x _i1, x _i2, x _i3, x _i4, x _i5), wherein, x _i1corresponding chilled water leaving water temperature T _1o, x _i2corresponding cooling water leaving water temperature T _2o, x _i3corresponding chilled water pump flow v ₁, x _i4corresponding cooling water pump flow v ₂, x _i5corresponding air mass flow Fair, setting Particle Swarm scale is s, evolutionary generation is t _max, to each particulate X _i(i=1,2 ..., s) carry out the initialization of position and speed, to i arbitrarily, j (j=1,2 ..., 5), all in its range of variables, obey equally distributed generation x _ijand v _ijinitialization value, and the local desired positions P to particulate _iwith overall desired positions P _ginitialize;

Second step, speed and the position of renewal particulate, formula is expressed as follows:

v _ij(t+1)＝ω·v _ij(t)+m ₁r ₁(t)(p _ij(t)-x _ij(t))+m ₂r ₂(t)(p _gj(t)-x _ij(t))；

(i＝1,2,…,s)(j＝1,2,…,5)

x _ij(t+1)＝x _ij(t)+v _ij(t+1)；(i＝1,2,…,s)(j＝1,2,…,5)

Wherein, v _ijand x (t) _ij(t) be respectively particulate t for time speed and position, ω is inertia weight, m ₁and m ₂for aceleration pulse, r ₁and r ₂for the random number between 0 and 1, P _i(t) be the local desired positions in t generation of particulate, P _g(t) be the t overall desired positions in generation, p _ij(t) be vectorial P _i(t) the j dimension element in, p _gj(t) be vectorial P _g(t) the j dimension element in;

The 3rd step, calculates the target function value f (X of each particulate in t+1 generation _i);

The 4th step, the local desired positions P of renewal particulate _iwith overall desired positions P _g, formula table is shown:

$P_i(t+1)=\left\{\begin{array}{cc}{P}_i\left(t\right)& f\left(X_i(t+1)\right)\&GreaterEqual;f\left(P_i\left(t\right)\right)\\ X_i(t+1)& f\left(X_i(t+1)\right)<f\left(P_i\left(t\right)\right)\end{array}\right.$

$P_g(t+1)=\left\{\begin{array}{cc}{P}_g\left(t\right)& f\left(X_i(t+1)\right)\&GreaterEqual;f\left(P_g\left(t\right)\right)\\ X_i(t+1)& f\left(X_i(t+1)\right)<f\left(P_g\left(t\right)\right)\end{array}\right.$

P wherein _i(t+1) be the local desired positions in t+1 generation of particulate, P _g(t+1) be the t+1 overall desired positions in generation, X _i(t+1) be that t+1 is for particulate;

The 5th step, if evolutionary generation t < is t _max, return to second step, otherwise finish and export overall desired positions P _g;

Step 4: according to the T obtaining _1o, T _2o, v ₁, v ₂, the best parameter group of Fair, controls in advance to each parametric variable, and when guaranteeing that moment to be measured arrives, air-conditioning system is in optimum duty.

2. air conditioner system energy saving optimization method as claimed in claim 1, the historical data of wherein collecting comprises room occupancy rate and the air-conditioning system load in a certain moment, and the outdoor mean temperature before the described a certain moment and air-conditioning system load.