CN115719294A - Method, system, electronic equipment and medium for indoor pedestrian flow evacuation control - Google Patents

Abstract

The invention discloses an indoor pedestrian flow evacuation control method, an indoor pedestrian flow evacuation control system, electronic equipment and an indoor pedestrian flow evacuation control medium, and relates to the field of pedestrian evacuation, wherein the method comprises the steps of firstly constructing a state equation and a measurement equation of outlet pedestrian density and a BP neural network for identifying a state function and an observation function and performing offline iterative training; solving a state function and an observation function according to the weight values of all layers in the BP neural network; predicting pedestrian flow evacuation density according to the solved function, and constructing an error online neural network for correcting a predicted value; and when the data state is abnormal, calculating the sum of the pedestrian flow evacuation density prediction value and the prediction error value as a pedestrian flow evacuation density prediction improvement value, and performing indoor pedestrian flow evacuation control according to the improvement value. The method can still ensure the accuracy of pedestrian flow evacuation density prediction under the condition of performance reduction or damage of the sensor, thereby improving the pedestrian flow evacuation efficiency.

Description

Method, system, electronic equipment and medium for indoor pedestrian flow evacuation control

technical field

The invention relates to the technical field of pedestrian evacuation, in particular to an indoor pedestrian flow evacuation control method, system, electronic equipment and medium.

Background technique

With the continuous development of society, public places such as entertainment facilities and large shopping malls in various parts of the city have increased dramatically. And above-mentioned public places often have the situation that crowd gathers highly, after disasters such as fire, earthquake take place, very easily produce stampede incident. In the event of an emergency, if the guide personnel can effectively evacuate, casualties will be greatly reduced. Therefore, crowd evacuation has become a hot research topic.

Comparing the current research status at home and abroad in China, path finding algorithms, cellular automata models, and reinforcement learning control strategies are widely used in the research topics of pedestrian evacuation. However, the prerequisite for the above algorithm to perform well in the process of crowd evacuation is the accurate measurement of sensors (such as cameras). If there are uncontrollable factors such as smoke around the sensor at this time, it may cause the sensor to measure the distribution of pedestrians inaccurately, thus affecting the working efficiency of the evacuation algorithm. In practical applications, the reduction in the efficiency of the evacuation algorithm may cause irreparable losses of life and property.

Contents of the invention

The purpose of the present invention is to provide an indoor pedestrian flow evacuation control method, system, electronic equipment and medium, so as to improve the pedestrian flow evacuation efficiency under the condition of sensor performance degradation or damage.

To achieve the above object, the present invention provides the following scheme:

In one aspect, the present invention provides a method for controlling indoor pedestrian flow evacuation, including:

Model the evolution rules of pedestrian density in indoor evacuation scenarios, and construct the state equation and measurement equation of exit pedestrian density;

Constructing a BP neural network for identifying state functions and observation functions in the state equation and measurement equation; the BP neural network includes an input layer, a hidden layer and an output layer;

Carry out off-line iterative training to described BP neural network, obtain the weight of each layer in BP neural network after training finishes;

Solve the state function and the observation function in the state equation and the measurement equation according to the weights of each layer in the BP neural network;

Predict the evacuation density of pedestrian flow according to the obtained state function and observation function, and obtain the predicted value of evacuation density of pedestrian flow;

Constructing an error online neural network for correcting the predicted value of the pedestrian flow evacuation density;

judging whether there is an abnormal data state according to the sensor state in the indoor evacuation scene;

If there is no data state abnormality, directly perform indoor pedestrian flow evacuation control according to the predicted value of the pedestrian flow evacuation density, and continuously train the error online neural network;

If there is an abnormality in the data state, then calculate the prediction error value according to the current trained error online neural network, and calculate the sum of the pedestrian flow evacuation density prediction value and the described prediction error value as the pedestrian flow evacuation density prediction improvement value;

Indoor pedestrian flow evacuation control is performed according to the improved value of the pedestrian flow evacuation density prediction.

Optionally, the step of modeling the pedestrian density evolution rule in the indoor evacuation scene, constructing the state equation and measurement equation of the exit pedestrian density, specifically includes:

Model the pedestrian density evolution rules in the indoor evacuation scene, construct the state equation X(k)=f(X(k-1))+q(k-1) and the measurement equation Z(k)= h(X(k))+r(k); wherein the indoor evacuation scene has multiple exits, and each exit is equipped with evacuation guide personnel and sensors; X(k)∈R ^n×n and X(k- 1) represent the state variables at time k and k-1 respectively; Z(k)∈R ^m×n represents the observed variables at time k; f(·) represents the nonlinear state function; h(·) represents the nonlinear observation function ; q(k-1)∈R ^n×n represents process noise; r(k)∈R ^m×n represents measurement noise.

Optionally, the construction of a BP neural network for identifying the state function and the observation function in the state equation and the measurement equation specifically includes:

Construct the BP neural network for identifying the nonlinear state function f(·) in the state equation, the BP neural network takes X(k) as input, Z(k) as output, Sigmoid function as activation function, function The expression is Z(k)=h(X(k))=W ₂ ^T sigmoid(W ₁ ^T X(k)); where W ₁ is the weight from the input layer to the hidden layer, and W ₂ is the weight from the hidden layer to the output layer weights;

Constructing a BP neural network for identifying the nonlinear observation function h(·) in the measurement equation, the BP neural network takes X(k-1) as input, X(k) as output, and Sigmoid function as activation function , the function expression is X(k)=f(X(k-1))=W ₂ ^T sigmoid(W ₁ ^T X(k-1)).

Optionally, the off-line iterative training of the BP neural network is carried out, and the weights of each layer in the BP neural network are obtained after training, specifically including:

Based on the pedestrian evacuation simulation experiment, the sample data used to train the BP neural network is obtained, and the sample data is divided into a training set and a verification set at a ratio of 7:3. At the same time, SGD is used as an optimizer to update weights, and MSE is used as a loss function. Iterative training is carried out with a fixed learning rate and training times, and the weights of each layer in the BP neural network are obtained after the training.

Optionally, the construction of an error online neural network for correcting the predicted value of pedestrian flow evacuation density specifically includes:

Based on the BP neural network, the error online neural network used to correct the predicted value of the evacuation density of the pedestrian flow is constructed; the input of the online neural network of the error is the predicted value of the evacuation density of the pedestrian flow, and the output is the predicted value of the evacuation density of the pedestrian flow. The error between the true values of the evacuation density of people, that is, the forecast error value.

Optionally, the judging whether there is an abnormal data state according to the sensor state in the indoor evacuation scene specifically includes:

If the sensor in the indoor evacuation scene has a performance degradation or a damaged state, it is determined that there is an abnormal data state;

If the sensor in the indoor evacuation scene does not have a performance degradation or a damaged state, it is determined that there is no abnormal data state.

Optionally, the indoor pedestrian flow evacuation control according to the improved value of the pedestrian flow evacuation density prediction includes:

进行室内行人流疏散控制；其中

表示室内疏散场景中k时刻第i个出口的行人流疏散目标密度；u_i(k)表示k时刻第i个出口的引导作用系数；当u_i(k)为1时打开第i个出口的引导作用，当u_i(k)为0时关闭第i个出口的引导作用。Substitute the improved value ρ _i (k) of pedestrian flow evacuation density prediction at the i-th exit at time k into the formula

Carry out indoor pedestrian flow evacuation control;

Indicates the pedestrian evacuation target density of the i-th exit at time k in the indoor evacuation scene; u _i (k) represents the guiding effect coefficient of the i-th exit at k time; when u _i (k) is 1, the Guidance function, when u _i (k) is 0, close the guidance function of the i-th exit.

On the other hand, the present invention also provides an indoor pedestrian flow evacuation control system, including:

The pedestrian density equation modeling module is used to model the evolution rules of pedestrian density in indoor evacuation scenarios, and construct the state equation and measurement equation of pedestrian density at exits;

A BP neural network building block, used to construct a BP neural network for identifying state functions and observation functions in the state equation and measurement equation; the BP neural network includes an input layer, a hidden layer and an output layer;

The off-line network iterative training module is used to carry out off-line iterative training to the BP neural network, and the training ends to obtain the weights of each layer in the BP neural network;

A state and observation function solving module, used to solve the state function and the observation function in the state equation and measurement equation according to the weights of each layer in the BP neural network;

The pedestrian flow evacuation density prediction module is used to predict the pedestrian flow evacuation density according to the solved state function and the observation function, and obtain the pedestrian flow evacuation density prediction value;

An error online neural network building block is used to construct an error online neural network for correcting the predicted value of the pedestrian flow evacuation density;

A data state abnormal judgment module, configured to judge whether there is a data state abnormality according to the sensor state in the indoor evacuation scene;

The error online neural network training module is used to directly perform indoor pedestrian flow evacuation control according to the predicted value of the pedestrian flow evacuation density if there is no abnormal data state, and continuously train the error online neural network;

The density prediction improved value calculation module is used to calculate the prediction error value according to the currently trained error online neural network if there is an abnormal data state, and calculate the sum of the pedestrian flow evacuation density prediction value and the prediction error value as the row Improvement value of crowd evacuation density prediction;

The indoor pedestrian flow evacuation control module is used to perform indoor pedestrian flow evacuation control according to the improved value of the pedestrian flow evacuation density prediction.

On the other hand, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and operable on the processor, and the processor implements the computer program when executing the computer program. The above-mentioned indoor pedestrian flow evacuation control method.

On the other hand, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed, the above-mentioned indoor pedestrian flow evacuation control method is realized.

According to the specific embodiments provided by the invention, the invention discloses the following technical effects:

The present invention provides an indoor pedestrian flow evacuation control method, system, electronic equipment and medium. The method includes: modeling the evolution rules of pedestrian density in an indoor evacuation scene, and constructing a state equation and a measurement equation of pedestrian density at an exit; Construction is used to identify the BP neural network of state function and observation function in described state equation and measurement equation; Described BP neural network comprises input layer, hidden layer and output layer; Described BP neural network is carried out off-line iterative training, training End to obtain the weights of each layer in the BP neural network; solve the state function and the observation function in the state equation and the measurement equation according to the weights of each layer in the BP neural network; perform the operation according to the state function and the observation function solved Prediction of the evacuation density of the pedestrian flow, obtaining the predicted value of the evacuation density of the pedestrian flow; constructing an error online neural network for correcting the predicted value of the evacuation density of the pedestrian flow; judging whether there is an abnormal data state according to the sensor state in the indoor evacuation scene ; If there is no data state abnormality, directly carry out indoor pedestrian flow evacuation control according to the predicted value of the pedestrian flow evacuation density, and continuously train the error online neural network; if there is an abnormal data state, then according to the currently trained error online neural network The network calculates the prediction error value, and calculates the sum of the pedestrian flow evacuation density prediction value and the prediction error value as the pedestrian flow evacuation density prediction improvement value; performs indoor pedestrian flow evacuation control according to the pedestrian flow evacuation density prediction improvement value. The method of the invention can still ensure the accuracy of pedestrian flow evacuation density prediction under the condition of sensor performance degradation or damage, thereby improving pedestrian flow evacuation efficiency.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some of the present invention. Embodiments, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without any creative effort.

Fig. 1 is the flow chart of a kind of indoor pedestrian flow evacuation control method of the present invention;

Fig. 2 is the technical roadmap of a kind of indoor pedestrian flow evacuation control method of the present invention;

FIG. 3 is a schematic diagram of an indoor evacuation scene provided by an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a BP neural network for nonlinear observation function identification provided by an embodiment of the present invention;

5 is a schematic structural diagram of a BP neural network for nonlinear state function identification provided by an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of an error online neural network provided by an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

The object of the present invention is to provide a method, system, electronic device and medium for indoor pedestrian flow evacuation control, and realize the prediction of pedestrian flow evacuation density in the field of view of sensors (such as cameras) on the basis of pedestrian flow evacuation density control simulation. During the evacuation process, the performance of the sensor is degraded or even damaged (for example, the camera is damaged due to a fire), and the pedestrian flow evacuation density control cannot work normally, so as to improve the pedestrian flow evacuation efficiency under the condition of sensor performance degradation or damage.

In order to make the above objects, features and advantages of the present invention more comprehensible, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

Fig. 1 is a flow chart of an indoor pedestrian flow evacuation control method according to the present invention; Fig. 2 is a technical roadmap of an indoor pedestrian flow evacuation control method according to the present invention. Referring to Fig. 1 and Fig. 2, a kind of indoor pedestrian flow evacuation control method of the present invention comprises:

Step 1: Model the evolution rules of pedestrian density in the indoor evacuation scene, and construct the state equation and measurement equation of exit pedestrian density.

Fig. 3 is a schematic diagram of an indoor evacuation scene provided by an embodiment of the present invention. Referring to Fig. 3, the evacuation scene targeted by the present invention is an indoor evacuation scene, such as a large shopping mall, playground, library, museum, etc., the indoor evacuation scene has multiple exits, and each exit is equipped with evacuation guides and sensors; This sensor is usually a camera. The indoor evacuation scene is gridded to form a space including N×N grids. In the embodiment shown in Fig. 3, there are 4 exits, and each exit is equipped with evacuation guide personnel. Through the camera at the exit, the evacuation guidance personnel can obtain the pedestrian density at the exit and adjust the guidance signal (open or close), so as to realize the control of the entire pedestrian flow evacuation guidance process.

The invention models the pedestrian density evolution rule in the indoor evacuation scene, and constructs the state equation and measurement equation of the exit pedestrian density as follows.

The state equation of exit pedestrian density is:

X(k)=f(X(k-1))+q(k-1) (1)

The measurement equation of exit pedestrian density is:

Z(k)=h(X(k))+r(k) (2)

where X(k)∈R ^n×n and X(k-1)∈R ^n×n represent the state variables at time k and k-1 respectively; Z(k)∈R ^m×n represents the observation at time k variable; f(·) represents the nonlinear state function; h(·) represents the nonlinear observation function; q(k-1)∈R ^n×n represents the process noise; r(k)∈R ^m×n represents the measurement noise.

where r(k) and q(k-1) are independent of each other, have a mean of 0 and satisfy the following equation:

E[r(k)]=0, E[q(k-1)]=0 (3)

E[r(k)r ^T (k)]=R(k), E[q(k-1)q ^T (k-1)]=Q(k-1) (4)

R(k)的取值为0.1。In the formula, E[.] represents the expectation; R(k) and Q(k-1) represent the covariance matrix of r(k) and q(k-1) respectively, and both are Gaussian white noise. Among them, the value of Q(k-1) is

The value of R(k) is 0.1.

In the embodiment shown in Figure 3, assuming that there is interference from unknown factors around the fourth door (that is, exit 4), resulting in an error in the pedestrian density at the exit collected by the sensor, the sensor (camera) of the fourth door detects The human flow density obtained corresponds to the observed variable Z(k), and the human flow density detected by the sensors of the other three doors corresponds to the state variable X(k):

Z(k)＝(people flow density detected by the fourth sensor) (6)

Since the state function f( ) and observation function h( ) in the above equation cannot be solved directly, the present invention adopts BP neural network for identification, and then obtains the nonlinear state function f( ) and nonlinear observation function h (·).

Step 2: Construct a BP neural network for identifying the state function and observation function in the state equation and measurement equation.

For the embodiment shown in Figure 3, in the state equation (1) and measurement equation (2) for constructing exit pedestrian density constructed in step 1, the state variable X(k-1)∈R ^3×1 at time k-1 , the state variable X(k)∈R ^3×1 at time k, and the observed variable Z(k)∈R ^1×1 at time k. Since the input and output of the two functions f(·) and h(·) are different, two BP neural networks need to be constructed to identify the nonlinear state function f(·) and the nonlinear observation function h(·) . Both BP neural networks include an input layer, an intermediate layer (usually a hidden layer) and an output layer.

The BP neural network structure used for the identification of the nonlinear observation function h( ) is shown in Figure 4, where dense_1_input:ImputLayer, dense_1:Dense and dense_2:Dense respectively represent the input layer, intermediate layer and output layer of the BP neural network; input and output respectively represent the input and output of a certain layer of BP neural network. In Figure 4, the flow density of the 1st, 2nd, and 3rd gates at time k is used as input, so the input of the input layer is 3-dimensional, and None represents the number of samples required to be input, so the input of the input layer can be written as (None ,3); the input layer converts (None,3) into (None,7) according to the connection weights between the input layer and the middle layer, and similarly the middle layer obtains the output (None,1) of the output layer through the connection weights, namely The traffic density of the fourth door at time k. That is to say, the BP neural network used for the identification of the nonlinear observation function h(·) in the present invention takes X(k) as input, Z(k) as output, and the Sigmoid function as the activation function of the BP neural network. Therefore, the function expression of the BP neural network is:

Z(k)=h(X(k))=W ₂ ^T sigmoid(W ₁ ^T X(k)) (7)

Where W ₁ is the weight from the input layer to the hidden layer, and W ₂ is the weight from the hidden layer to the output layer; the Sigmoid function will nonlinearize the calculation results of each neuron in the neural network linearly weighted, endowing the neural network with nonlinear Mapping ability, the expression of its function is

The structure of the BP neural network used for the identification of the nonlinear state function f(·) is shown in Figure 5. The BP neural network takes the flow density of the 1st, 2nd, and 3rd gates at time k-1 as input, and the 1st gate at time k , 2,3 The traffic density of the gates is used as the output, so the difference between the BP neural network structure shown in Figure 5 and Figure 4 is that in the output layer dense_2:Dense of the BP neural network, its output dimension is 3-dimensional, that is, the input The layer converts (None, 3) into (None, 7) according to the connection weights between the input layer and the intermediate layer. Similarly, the output of the output layer obtained by the intermediate layer through the connection weight is (None, 3), that is, the first , 2,3 door traffic density. That is to say, the BP neural network used for the identification of the nonlinear state function f(·) in the present invention takes X(k-1) as input, X(k) as output, and the Sigmoid function as the activation function of the BP neural network. Therefore, the function expression of the BP neural network is:

X(k)=f(X(k-1))=W ₂ ^T sigmoid(W ₁ ^T X(k-1)) (8)

Where W ₁ is the weight from the input layer to the hidden layer, and W ₂ is the weight from the hidden layer to the output layer.

Step 3: Perform off-line iterative training on the BP neural network, and obtain the weights of each layer in the BP neural network after training.

On the basis of the BP neural network constructed in step 2, the sample data used to train the BP neural network is obtained based on the pedestrian evacuation simulation experiment, and the sample data is divided into a training set and a verification set in a ratio of 7:3, and SGD is used as The optimizer updates the weights, MSE is used as the loss function, and the learning rate and training times are set for iterative training. After the training, the weights W ₁ and W ₂ of each layer in the BP neural network can be obtained.

Specifically, a pedestrian evacuation simulation experiment based on the existing indoor multi-exit pedestrian flow simulation method is carried out. When the pedestrian evacuation simulation experiment is running, the pedestrian flow density data measured by the four gates at each moment during the pedestrian evacuation process will be recorded, and then random Take 1000 times of simulation results and save them, so as to obtain sample data for training the BP neural network of the present invention. Divide the sample data into a training set and a validation set at a ratio of 7:3, and at the same time use SGD as the optimizer to update the weight, MSE as the loss function, set the learning rate and the number of training times for iterative training. After the BP neural network is trained, the weights of each layer can be obtained.

Among them, the stochastic gradient descent strategy (SGD) adjusts the weight of the BP neural network in the direction of the negative gradient of the target gradient. For an output node:

Among them, η represents the learning rate, and the value is 0.01; E _k is the mean square error MSE; XW _ij represents the connection weight between the i-th layer and the j-th layer of the BP neural network, and ΔW _ij is the connection weight XW _ij in the training process of updates.

The connection weights and thresholds between the hidden layer and the input layer need to be adjusted as follows:

The calculated connection weight XW _ij is a scalar; W ₁₂ is the connection weight between the input layer and the middle layer, and it will be used as W ₁ in formulas (7) and (8) after training; W ₂₃ is is the connection weight between the middle layer and the output layer, and it will be used as W ₂ in the formulas (7) and (8) after the training. Both W ₁ and W ₂ are in matrix form.

In addition, in order to make the actual error and the prediction error with the network as small as possible, it is necessary to iteratively adjust the weights of each layer of the BP neural network according to the stochastic gradient descent strategy, so that the value of the loss function is continuously reduced.

Step 4: Solve the state function and observation function in the state equation and measurement equation according to the weight values of each layer in the BP neural network.

中，针对状态函数和观测函数分别为公式(7)和(8)中，即可得到非线性状态函数f(·)以及非线性观测函数h(·)。式中W₁为输入层到隐藏层的连接权值，W₂为隐藏层到输出层的连接权值，input为BP神经网络的输入，output为BP神经网络的输出。Substitute the weights W ₁ and W ₂ of each layer obtained from training into the transfer function of the corresponding BP neural network

, for the state function and the observation function are formulas (7) and (8) respectively, the nonlinear state function f(·) and the nonlinear observation function h(·) can be obtained. In the formula, W ₁ is the connection weight from the input layer to the hidden layer, W ₂ is the connection weight from the hidden layer to the output layer, input is the input of the BP neural network, and output is the output of the BP neural network.

After obtaining f(·) and h(·), the real-time prediction of the pedestrian flow density at a certain exit at each time can be carried out according to the formula (1) and formula (2), and the predicted value of the pedestrian flow evacuation density of the corresponding exit can be obtained .

Step 5: Predict the evacuation density of pedestrian flow according to the obtained state function and observation function, and obtain the predicted value of evacuation density of pedestrian flow.

According to the obtained nonlinear state function f( ) and nonlinear observation function h( ) to predict the evacuation density of pedestrian flow, the prediction process is as follows:

1) Calculate the prior estimate X(k|k-1); specifically, use the state variable X(k-1) at time k-1 into formula (11) to calculate the prior estimate of the state variable X(k) at time k X(k|k-1):

X(k|k-1)=f(X(k-1)) (11)

2) According to X(k|k-1), predict the pedestrian flow evacuation density prediction value Z ₄ (k) of the fourth exit at time k:

Z ₄ (k)=H(k)X(k|k-1) (12)

where H(k) is the Jacobian matrix of h(X(k|k-1)).

3) Calculate the covariance matrix P(k|k-1):

P(k|k-1)＝F(k|k-1)P(k-1) ^FT (k,k-1)+Q(k-1) (13)

Where F(k|k-1) is the Jacobian matrix of f(X(k-1)); P(k-1) represents the covariance matrix at time k-1.

4) Calculate the gain K(k), which is used to reflect the confidence of the predicted value and the observed value:

K(k)=P(k|k-1)HT( ^k )[H(k)P(k|k-1) ^HT (k)+R(k)] ^-1 (14)

5) Combined with the gain K(k), calculate the posterior estimate X(k), that is, modify X(k|k-1):

X(k)=X(k|k-1)+K(k)[Z(k)-h(X(k|k-1))] (15)

6) Calculate the covariance matrix P(k):

P(k)=[I-K(k)H(k)]P(k|k-1) (16)

In the formula, I is the unit matrix; F(k|k-1) is the Jacobian matrix of f(X(k-1)); H(k) is the Jacobian matrix of h(X(k|k-1)) . According to formulas (11) to (16), it is possible to recursively predict the flow density of people at a certain exit at each moment, for example, the predicted value Z ₄ (k) of pedestrian flow evacuation density at the fourth exit at time k and the first, The predicted value X(k) of pedestrian flow evacuation density at 2 and 3 exits.

Step 6: constructing an error online neural network for correcting the predicted value of the pedestrian flow evacuation density.

Step 5 can recursively predict the flow density of people at a certain exit at each moment, but as time continues to recur, the prediction error may continue to increase with the accumulation of time. On the other hand, if the sensors in the indoor evacuation scene have performance degradation or damage, there will be significant errors in the traffic density predicted based on the data collected by the sensors. In view of this problem, the present invention builds an error online neural network for correcting the predicted value of the pedestrian flow evacuation density based on a BP neural network, and corrects the predicted value by constructing an error online neural network. Its network structure diagram is shown in Figure 6 Show. In Figure 6, dense_1_input:ImputLayer, dense_1:Dense and dense_2:Dense respectively represent the input layer, intermediate layer and output layer of the BP neural network; input and output respectively represent the input and output of a certain layer of the BP neural network. In Figure 6, the flow density of the 1st, 2nd, and 3rd gates at time k is used as input, so the input of the input layer is 3-dimensional, and None represents the number of samples required to be input, so the input of the input layer can be written as (None ,3); the input layer converts (None,3) into (None,7) according to the connection weights between the input layer and the middle layer, and similarly the middle layer obtains the output (None,1) of the output layer through the connection weights, that is The error ΔZ(k) between the predicted value Z(k) of pedestrian flow evacuation density at time k and the real value _Ztrue (k) of pedestrian flow evacuation density at time k.

That is, the input of the error online neural network constructed by the present invention is the predicted value of pedestrian flow evacuation density, which is Z ₄ (k) calculated by formula (12) in this embodiment; the output is the predicted value of pedestrian flow evacuation density and pedestrian flow The error between the true values of the evacuation density, that is, the prediction error value ΔZ(k). Among them, the real value of pedestrian evacuation density _Ztrue (k) at time k adopts the pedestrian flow density data of the fourth door at time k during the pedestrian evacuation process recorded during the pedestrian evacuation simulation experiment. When there is no abnormality in the data state, Z ₄ (k)= _Ztrue (k); when there is an abnormality in the data state, _Ztrue (k)=Z ₄ (k)+ΔZ(k).

Step 7: Judging whether there is an abnormal data state according to the sensor state in the indoor evacuation scene.

The existing indoor multi-exit pedestrian flow simulation method needs to have high accuracy of the pedestrian flow density information obtained by the sensor to work normally. However, in an actual evacuation scene, such as a fire emergency evacuation scene, the performance of the sensor may be degraded due to smoke occlusion or the sensor may be damaged due to the fire. Inaccurate situation. The occurrence of the above-mentioned problems leads to the inability of the existing methods to continue to work normally, thereby affecting the evacuation efficiency. In order to solve this problem, the present invention proposes to predict and correct the pedestrian density data acquired by the camera to improve the accuracy of the data, and then perform indoor pedestrian flow evacuation control according to the corrected pedestrian flow evacuation density prediction improvement value, thereby reducing data errors Influence on the guidance control system.

Specifically, if the sensor in the indoor evacuation scene has a performance degradation or a damaged state, it is determined that there is an abnormal data state; if the sensor in the indoor evacuation scene does not have a performance degradation or a damaged state, then it is determined that there is no abnormal data state .

Step 8: If there is no abnormality in the data state, perform indoor pedestrian flow evacuation control directly according to the predicted value of the pedestrian flow evacuation density, and continuously train the error online neural network.

If the data collected by the sensor is in a normal state at present, the error online neural network is put into a training state, and the error online neural network is continuously trained. At this time, the indoor pedestrian flow evacuation control can be carried out directly according to the predicted value of the pedestrian flow evacuation density, and the control algorithm is as follows:

时，对应门的引导作用系数u_i(k)会被调整为1，从而打开引导作用。反之，则对应门的引导作用系数u_i(k)会被调整为0，从而关闭引导作用。However, when it is judged that there is no data state abnormality, the predicted value Z _i (k) of the pedestrian flow evacuation density of the i-th exit at k time (Z ₄ (k) in the embodiment of the present invention) can be used as ρ _i (k) and substituted into Equation (17) performs indoor pedestrian flow evacuation control. Among them, ρ _i ^aim represents the pedestrian flow evacuation target density of the i-th exit at time k in the indoor evacuation scene; u _i (k) represents the guiding coefficient of the i-th exit at k time; when u _i (k) is 1, open the first The guidance function of the i exit, when u _i (k) is 0, the guidance function of the i-th exit is turned off. That is, when the exit pedestrian density ρ _i (k) obtained by the evacuation guide personnel at the i-th exit at time k is less than or equal to the set target density

When , the guiding effect coefficient u _i (k) of the corresponding door will be adjusted to 1, thus turning on the guiding effect. On the contrary, the guiding effect coefficient u _i (k) of the corresponding door will be adjusted to 0, thus turning off the guiding effect.

Step 9: If there is an abnormal data state, calculate the prediction error value according to the currently trained error online neural network, and calculate the sum of the pedestrian flow evacuation density prediction value and the prediction error value as the pedestrian flow evacuation density prediction improvement value .

If the current data collected by the sensor is in an abnormal state, the error online neural network enters the use state at this time, and the predicted error value ΔZ(k) is calculated according to the currently trained error online neural network. The network prediction value and the prediction error value obtained by the error online neural network are combined to obtain an improved result. That is, the sum of the predicted pedestrian evacuation density value Z ₄ (k) and the predicted error value ΔZ(k) is calculated according to the following formula (18) as the pedestrian flow evacuation density predicted _improved value Z(k):

Z _change (k) = Z ₄ (k) + ΔZ (k) (18)

Step 10: Perform indoor pedestrian flow evacuation control according to the improved value of the pedestrian flow evacuation density prediction.

When it is judged that there is an abnormality in the data state, the improved value Z of pedestrian flow evacuation density prediction at the i-th exit at time k _{is changed to} (k) as ρ _i (k), and is substituted into the above formula (17) for indoor pedestrian flow evacuation control. In the embodiment of the present invention, when u ₄ (k) is 1, the guiding function of the fourth outlet is turned on, and when u ₄ (k) is 0, the guiding function of the fourth outlet is turned off. The results show that after the pedestrian density data acquired by the camera is corrected by the method of the present invention, the effect of indoor pedestrian flow evacuation control has been significantly improved.

It can be seen that the method of the present invention firstly models the pedestrian density evolution rule in the indoor evacuation scene, and constructs the nonlinear state function and nonlinear observation function expression of the exit pedestrian density; secondly, the BP neural network identification is used to solve the nonlinear state function and The parameters of the nonlinear observation function solve the problem that the function parameters cannot be directly solved by derivation; again, the prediction value is optimized and adjusted in real time through the error online neural network to further improve the prediction accuracy; finally, the real-time prediction improvement value is used for pedestrian flow Evacuation control improves evacuation efficiency in the event of sensor degradation or damage. The method of the present invention realizes the prediction of the pedestrian evacuation density in the field of view of the sensor (such as a camera) on the basis of the existing pedestrian flow evacuation density control simulation, thereby solving the problem of sensor performance degradation or even damage during the evacuation process (such as camera damage caused by fire) After that, the pedestrian flow evacuation density control does not work properly.

Based on the method provided by the present invention, the present invention also provides an indoor pedestrian flow evacuation control system, including:

The pedestrian density equation modeling module is used to model the evolution rules of pedestrian density in indoor evacuation scenarios, and construct the state equation and measurement equation of pedestrian density at exits;

A BP neural network building block, used to construct a BP neural network for identifying state functions and observation functions in the state equation and measurement equation; the BP neural network includes an input layer, a hidden layer and an output layer;

The off-line network iterative training module is used to carry out off-line iterative training to the BP neural network, and the training ends to obtain the weights of each layer in the BP neural network;

A state and observation function solving module, used to solve the state function and the observation function in the state equation and measurement equation according to the weights of each layer in the BP neural network;

The pedestrian flow evacuation density prediction module is used to predict the pedestrian flow evacuation density according to the solved state function and the observation function, and obtain the pedestrian flow evacuation density prediction value;

An error online neural network building block is used to construct an error online neural network for correcting the predicted value of the pedestrian flow evacuation density;

A data state abnormal judgment module, configured to judge whether there is a data state abnormality according to the sensor state in the indoor evacuation scene;

The error online neural network training module is used to directly perform indoor pedestrian flow evacuation control according to the predicted value of the pedestrian flow evacuation density if there is no abnormal data state, and continuously train the error online neural network;

The density prediction improved value calculation module is used to calculate the prediction error value according to the currently trained error online neural network if there is an abnormal data state, and calculate the sum of the pedestrian flow evacuation density prediction value and the prediction error value as the row Improvement value of crowd evacuation density prediction;

The indoor pedestrian flow evacuation control module is used to perform indoor pedestrian flow evacuation control according to the improved value of the pedestrian flow evacuation density prediction.

On the whole, the method and system of the present invention use the modeling method to predict and correct the pedestrian density data obtained by the camera, thereby improving the accuracy of the data; and then carry out indoor pedestrian flow evacuation control according to the corrected data, thereby reducing data errors. The impact of the control system, thereby improving the evacuation efficiency in the event of sensor degradation or damage. The invention identifies the nonlinear state function and the nonlinear observation function training set through the neural network identification method, thereby solving the problem that the model of the pedestrian flow evacuation system is complicated and difficult to directly solve. The present invention also adopts an offline method for system identification. Since the identification data obtained in real time is less in the experiment, the offline identification can make full use of historical data and effectively improve the training efficiency of identification. The present invention also uses the error online neural network to correct the predicted data. If only relying on the continuous recursion of the model, the error will continue to enlarge with the accumulation of time. Correction, thereby improving the robustness of the method of the present invention in practical applications.

Further, the present invention also provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus. Wherein, the processor, the communication interface, and the memory complete the mutual communication through the communication bus. The processor can call the computer program in the memory to execute the indoor pedestrian flow evacuation control method.

In addition, when the above-mentioned computer program in the memory is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, server or network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media capable of storing program codes such as U disk, mobile hard disk, read-only memory, random access memory, magnetic disk or optical disk.

Further, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed, the above-mentioned indoor pedestrian flow evacuation control method can be realized.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other. As for the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for the related information, please refer to the description of the method part.

In this paper, specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; meanwhile, for those of ordinary skill in the art, according to the present invention Thoughts, there will be changes in specific implementation methods and application ranges. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims (10)

1. An indoor pedestrian flow evacuation control method is characterized by comprising the following steps:

modeling a pedestrian density evolution rule in an indoor evacuation scene, and constructing a state equation and a measurement equation of the pedestrian density at an outlet;

constructing a BP neural network for identifying a state function and an observation function in the state equation and the measurement equation; the BP neural network comprises an input layer, a hidden layer and an output layer;

performing offline iterative training on the BP neural network, and obtaining weights of all layers in the BP neural network after the training is finished;

solving a state function and an observation function in the state equation and the measurement equation according to the weight values of all layers in the BP neural network;

predicting pedestrian flow evacuation density according to the solved state function and the observation function to obtain a pedestrian flow evacuation density prediction value;

constructing an error online neural network for correcting the predicted pedestrian flow evacuation density value;

judging whether data state abnormity exists according to the state of the sensor in the indoor evacuation scene;

if the data state is not abnormal, indoor pedestrian flow evacuation control is directly carried out according to the pedestrian flow evacuation density predicted value, and the error online neural network is continuously trained;

if the data state is abnormal, calculating a prediction error value according to a currently trained error online neural network, and calculating the sum of the pedestrian flow evacuation density prediction value and the prediction error value to serve as a pedestrian flow evacuation density prediction improvement value;

and carrying out indoor pedestrian flow evacuation control according to the pedestrian flow evacuation density prediction improvement value.

2. The indoor pedestrian flow evacuation control method according to claim 1, wherein modeling pedestrian density evolution rules in an indoor evacuation scenario to construct an equation of state and a measurement equation of exit pedestrian density specifically comprises:

modeling a pedestrian density evolution rule in an indoor evacuation scene, and constructing a state equation X (k) = f (X (k-1)) + q (k-1) and a measurement equation Z (k) = h (X (k)) + r (k) of the exit pedestrian density; wherein the indoor evacuation scenario has a plurality of exits, and each exit is equipped with an evacuation guidance person and a sensor; x (k) is belonged to R ^n×n And X (k-1) represents state variables at time k and time k-1, respectively; z (k) is belonged to R ^m×n An observed variable representing time k; f (-) represents a nonlinear state function; h (-) represents a nonlinear observation function; q (k-1) is epsilon to R ^n×n Representing process noise; r (k) is belonged to R ^m×n Representing the measurement noise.

3. The indoor pedestrian flow evacuation control method according to claim 2, wherein the constructing a BP neural network for identifying a state function and an observation function in the state equation and the measurement equation specifically comprises:

constructed forIdentifying a BP neural network of a nonlinear state function f (·) in the state equation, wherein the BP neural network takes X (k) as input, Z (k) as output and a Sigmoid function as an activation function, and the function expression is Z (k) = h (X (k)) = W ₂ ^T sigmoid(W ₁ ^T X (k)); wherein W ₁ As a weight of the input layer to the hidden layer, W ₂ The weight from the hidden layer to the output layer;

constructing a BP neural network for identifying a nonlinear observation function h (·) in the measurement equation, wherein the BP neural network takes X (k-1) as an input, X (k) as an output and a Sigmoid function as an activation function, and the function expression is X (k) = f (X (k-1)) = W ₂ ^T sigmoid(W ₁ ^T X(k-1))。

4. The indoor pedestrian flow evacuation control method according to claim 3, wherein the performing offline iterative training on the BP neural network, and obtaining weights of each layer in the BP neural network after the training is finished, specifically comprises:

obtaining sample data for training the BP neural network based on a pedestrian evacuation simulation experiment, and calculating the sample data by using a formula of 7: and 3, dividing the ratio into a training set and a verification set, updating the weight by taking SGD as an optimizer, taking MSE as a loss function, setting the learning rate and the training times to carry out iterative training, and obtaining the weight of each layer in the BP neural network after the training is finished.

5. The indoor pedestrian flow evacuation control method according to claim 4, wherein the constructing an error online neural network for correcting the predicted pedestrian flow evacuation density value specifically comprises:

constructing an error online neural network for correcting the pedestrian flow evacuation density predicted value based on the BP neural network; the input of the error online neural network is a pedestrian flow evacuation density predicted value, and the output is an error between the pedestrian flow evacuation density predicted value and a pedestrian flow evacuation density true value, namely a prediction error value.

6. The indoor pedestrian flow evacuation control method according to claim 5, wherein the determining whether there is a data state anomaly according to the sensor state in the indoor evacuation scene specifically includes:

if the sensors in the indoor evacuation scene have performance degradation or damage states, determining that data state abnormity exists;

and if the sensors in the indoor evacuation scene do not have performance degradation or damage states, determining that no data state abnormity exists.

7. The indoor pedestrian stream evacuation control method according to claim 6, wherein the indoor pedestrian stream evacuation control according to the predicted pedestrian stream evacuation density improvement value specifically includes:

predicting and improving value rho of pedestrian stream evacuation density of ith exit at moment k _i (k) Substitution formula

Carrying out indoor pedestrian flow evacuation control; wherein

Representing the pedestrian flow evacuation target density of the ith exit at the k moment in the indoor evacuation scene; u. of _i (k) The guiding action coefficient of the ith outlet at the k moment is represented; when u is _i (k) Opening the guide of the ith outlet at 1, when u _i (k) At 0, the guiding action of the ith outlet is closed.

8. An indoor pedestrian flow evacuation control system, comprising:

the pedestrian density equation modeling module is used for modeling a pedestrian density evolution rule in an indoor evacuation scene and constructing a state equation and a measurement equation of the pedestrian density at an outlet;

the BP neural network construction module is used for constructing a BP neural network used for identifying state functions and observation functions in the state equation and the measurement equation; the BP neural network comprises an input layer, a hidden layer and an output layer;

the offline network iterative training module is used for performing offline iterative training on the BP neural network, and obtaining weights of all layers in the BP neural network after the training is finished;

the state and observation function solving module is used for solving state functions and observation functions in the state equation and the measurement equation according to the weights of all layers in the BP neural network;

the pedestrian flow evacuation density prediction module is used for predicting the pedestrian flow evacuation density according to the solved state function and the observation function to obtain a pedestrian flow evacuation density prediction value;

the error online neural network construction module is used for constructing an error online neural network for correcting the pedestrian flow evacuation density predicted value;

the data state abnormity judging module is used for judging whether data state abnormity exists according to the state of the sensor in the indoor evacuation scene;

the error online neural network training module is used for directly carrying out indoor pedestrian flow evacuation control according to the pedestrian flow evacuation density predicted value and continuously training the error online neural network if the data state abnormity does not exist;

the density prediction improvement value calculation module is used for calculating a prediction error value according to a currently trained error online neural network if the data state is abnormal, and calculating the sum of the pedestrian flow evacuation density prediction value and the prediction error value to be used as a pedestrian flow evacuation density prediction improvement value;

and the indoor pedestrian flow evacuation control module is used for carrying out indoor pedestrian flow evacuation control according to the pedestrian flow evacuation density prediction improvement value.

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the indoor pedestrian flow evacuation control method according to any one of claims 1 to 7 when executing the computer program.

10. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed implements the indoor pedestrian flow evacuation control method of any one of claims 1 to 7.

Images