Elevator fault prediction method based on Elman neural network optimized by improved slime mold algorithm

Abstract

An elevator fault prediction method based on an improved slime mold algorithm to optimize the Elman neural network, using sensors to collect elevator operation data, uploading data to the background through a WIFI module for data processing, the background uses an improved slime mold algorithm to optimize the Elman neural network model, the slime mold algorithm is optimized by sine-tent chaotic mapping and adding greedy steps, and the optimization result with higher precision is output to the Elman neural network model, thereby optimizing the weight and threshold of the Elman model, and finally outputting the prediction result; after the operation data of the elevator is pre-processed and input into the optimized Elman model, the possible faults that may occur during the operation of the elevator in the future can be predicted; after sine-tent chaotic mapping and adding greedy steps for improvement, the improved slime mold algorithm has better optimization ability than the original algorithm, thereby making the final output better prediction results. The present invention can predict faults in advance, so that elevator maintenance personnel can more conveniently maintain the elevator and reduce the occurrence of accidents.

Description

Technical Field

[0001] The present invention relates to the technical field of elevator failure, and in particular to an elevator failure prediction method based on an improved slime mold algorithm to optimize an Elman neural network. Background Art

[0002] With the development of science and technology, elevators have become one of the indispensable operating equipment in high-rise buildings, mainly used for people's riding or cargo transportation. Elevators are a commonly used vertical riding tool, consisting of a traction system, a guide system, a car, a door system, a weight balance system, an electric traction system, an electrical control system, and a safety protection system. Although elevators greatly facilitate people's lives, due to their complex component systems, various elevator failures emerge in an endless stream, such as: elevator door closing failure, top rush, squatting, squeezing passengers, shearing, etc. Many of these failures even endanger the lives of passengers. In response to these problems, a variety of elevators with higher safety have emerged, mainly using cameras and multiple sensors to determine that the elevator has a fault and then repair or rescue it, but this method often has the problem of time lag. Summary of the invention

[0003] In response to the above technical problems, this technical solution utilizes the good data analysis ability of the Elman neural network model and the optimization ability of the slime mold algorithm, and provides an elevator fault prediction method based on the improved slime mold algorithm to optimize the Elman neural network. The various operating data of the monitored elevator are optimized through the improved slime mold algorithm, and the optimization results are input into the Elman model to optimize the weights and thresholds of the Elman model, so that the model has better prediction accuracy to predict the possibility of elevator failure in the future. It solves the problem of time lag for maintenance personnel or rescue personnel to obtain information after an elevator failure, and can better and faster prevent the occurrence of major accidents; it can effectively solve the above problems.

[0004] The present invention is achieved through the following technical solutions:

[0005] An elevator fault prediction method based on an improved slime mold algorithm to optimize the Elman neural network uses sensors to collect elevator operation data, and uploads the data to the background for data processing through a WIFI module. The background data processing uses an Elman neural network model optimized by an improved slime mold algorithm; the slime mold algorithm is optimized by sine-tent chaotic mapping and adding a new greedy step optimization, and outputs the optimization result with higher accuracy to the Elman neural network model, thereby optimizing the weights and thresholds of the Elman model, and finally outputting the prediction result; the specific operation steps include:

[0006] Step 1: Use sensors to collect various elevator operation data;

[0007] Step 2: Upload the collected data to the background for data preprocessing;

[0008] Step 3: For the data preprocessing in step 2, the collected data is normalized, the data is uniformly classified as [0,1], and input into the Elman neural network model;

[0009] Step 4: weights and thresholds of the initial Elman model;

[0010] Step 5: Use the slime mold algorithm to obtain the optimal value;

[0011] Step 6: Determine whether the end condition is met. If the condition is met, assign the optimal value to the Elman neural network; if not, repeat step 5;

[0012] Step 7: According to the output results of the Elman neural network, determine whether the elevator will have a corresponding elevator failure.

[0013] Furthermore, the slime mold algorithm is optimized by sine-tent chaotic mapping and adding a new greedy step. The formula of the added greedy step is as follows:

[0014]

[0015] Formula (1) allows the slime mold individual to search near the current optimal value, compares the search result with the fitness of the current individual, and if the fitness value of the new individual is better than that of the current individual, updates the position to the new individual, otherwise retains the original position;

[0016] Among them, x n1 is a new individual, x best is the current optimal individual, y1 is the oscillation decay function, randA is a uniform random number in [-a, a], x r is a random individual in the entire population, x i is the current individual, t is the independent variable, and F is the fitness value.

[0017] Furthermore, the slime mold algorithm is optimized by sine-tent chaotic mapping, so that the slime mold algorithm increases its global search capability when updating the position. The formula for adding sine-tent chaotic mapping optimization to the slime mold algorithm is as follows:

[0018]

[0019] x n2 =x best +k n *(ω n x1-x2) (3)

[0020] In formula (2), k n is the sine-tent mapping in [0,1]; Formula (3) is the individual update formula of the slime mold algorithm after sine-tent chaos optimization; the individual update formula after adding chaos mapping has better global search performance and effectively jumps out of the local optimum; in formula (3), k n is a sine-tent mapping in [0,1], ω1 is the weight, x best is the current optimal individual, x1 and x2 are two random individuals.

[0021] Furthermore, the formula for the normalization process in step 2 is as follows:

[0022]

[0023] In formula (4), x is the value of the current position, x' is the value after normalization, and x min is the minimum value in the data set, x max is the maximum value in the data set, c and d are the maximum and minimum values ​​after normalization, and the values ​​of c and d in formula (4) are [0,1].

[0024] Furthermore, the slime mold algorithm described in step 5 obtains the optimal value. Through the three parts of sorting, weight updating, and position updating, the slime mold algorithm can obtain an optimal individual, and input the optimal individual into the Elman model to optimize the weight or threshold of the model; the specific operation methods of the three parts of sorting, weight updating, and position updating are as follows:

[0025] S1. Sorting: sorting the fitness of the population to prepare for subsequent calculations;

[0026] S2. Weight update: When updating weights, different individuals have different update methods. The half of the population with better fitness uses formula (5), and the half with worse fitness uses formula (6);

[0027]

[0028] In formula (5) and formula (6), x best is the current optimal individual, x w is the worst fitness value in this iteration. The half of the population with better fitness has a larger weight after the weight update formula. The better the fitness, the closer the weight of the individual is to 1, and vice versa.

[0029] S3. Position update: Update the position according to the formula. The update formula is:

[0030]

[0031] p = tanh(|F(x best )-F(x i )|) (9)

[0032] x n3 =rand*(x max -x min )+x min rand1≤z (10)

[0033] x n4 =x best +randA*(ω1x1-x2)

[0034] rand1>z,rand2≤p (11)

[0035] x n5 =randB*x i rand1>z,rand2>p (12)

[0036] Formula (7) and Formula (8) are parameter update formulas. As the number of iterations finally approaches 0, Formula (9) is a threshold formula, which is used to decide whether to use Formula (10) or Formula (11) when updating the position. In Formulas (7) and (8), a and b are values ​​that gradually iterate to 0, which are used to control the upper and lower limits of randA and randB, p is the judgment threshold, t is the current iteration number, and T is the maximum iteration number.

[0037] Formula (10)(11)(12) is a specific position update formula. Formula (10) is used to obtain a random position; Formula (11) searches near the current optimal position; Formula (12) allows the individual to converge to 0;

[0038] In formulas (10), (11), and (12), x n3 、x n4 、x n5 is the latest position of the individual, rand1 and rand2 are random numbers in [0,1], z is a constant with a value of 0.03; randA is a uniform random number in [-a,a], and randB is a uniform random number in [-b,b].

[0039] Furthermore, the elevator operation data described in step 1 include: elevator shaft temperature, traction machine vibration, traction machine temperature, car vibration, elevator speed, elevator acceleration, elevator door opening and closing time, elevator door vibration, etc., represented by U(1)-U(8).

[0040] Furthermore, the elevator failure described in step 7 can be divided into an elevator car operation failure and an elevator door failure, which are represented by Y(1) and Y(2).

[0041] Furthermore, the Elman neural network is a local feedback recursive neural network, which belongs to a nonlinear state space model. The topological structure of Elman can be divided into four layers: input layer, hidden layer, receiving layer and output layer.

[0042] Furthermore, the function of the receiving layer is to receive the output of the hidden layer and pass the information corresponding to the previous time to the hidden layer in the form of feedback to establish a local ring structure; since the connection layer units have a delayed memory effect on the characteristics contained therein, in the process of elevator fault prediction, the output value of the Elman neural network is more in line with the actual development trend of the data.

[0043] Furthermore, the mathematical expression of the Elman neural network is:

[0044] C(τ)=X(j-1) (13)

[0045] X(j)=f(ω1*U(j)+ω2*C(j)) (14)

[0046] Y(j)=g(ω3*X(j)) (15)

[0047] In formulas (13)(14)(15), C(j) is the feedback state vector; Y(j) is the output node vector; X(j) is the intermediate layer node unit vector; U(j) is the input vector; ω3 is the connection weight from the intermediate layer to the output layer; ω1 is the connection weight from the input layer to the intermediate layer; ω2 is the connection weight from the receiving layer to the intermediate layer; g() is the transfer function of the output neuron, which is a linear combination of the intermediate layer output; f() is the transfer function of the intermediate layer neuron, and the S function is often used.

[0048] Beneficial Effects

[0049] The elevator fault prediction method based on the improved slime mold algorithm to optimize the Elman neural network proposed in the present invention has the following beneficial effects compared with the prior art:

[0050] This technical solution utilizes the good data analysis ability of the Elman neural network model and the optimization ability of the slime mold algorithm. Through the various operating data of the monitored elevator, the improved slime mold algorithm is used for optimization, and the optimization results are input into the Elman model to optimize the weights and thresholds of the Elman model, so that the model has better prediction accuracy to predict the possibility of elevator failure in the future, which solves the problem of time lag for maintenance personnel or rescue personnel to obtain information after an elevator failure occurs. It can better and faster prevent the occurrence of major accidents.

[0051] This technical solution uses sensors to collect elevator operation data, and uploads the data to the background through the WIFI module for data processing. The background data processing uses the Elman neural network model optimized by the improved slime mold algorithm. The slime mold algorithm is optimized by sine-tent chaotic mapping and adding new greedy steps to output the optimization results with higher precision to the Elman neural network model, thereby optimizing the weights and thresholds of the Elman model, and finally outputting the prediction results. After the elevator operation data collected by the sensor is pre-processed and input into the optimized Elman model, the possible faults that may occur during the operation of the elevator in the future are predicted. After the improvement of the two methods of sine-tent chaotic mapping and adding greedy steps, the improved slime mold algorithm has better optimization ability than the original algorithm, and thus outputs better prediction results in the end. This method predicts faults in advance, making it easier for elevator maintenance personnel to maintain the elevator and reduce the occurrence of accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the fault prediction process of the present invention.

[0053] Figure 2 It is a schematic diagram of the Elman neural network structure in the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Under the premise of not departing from the design concept of the present invention, various modifications and improvements made by ordinary persons in the art to the technical solutions of the present invention should all fall within the protection scope of the present invention.

[0055] Embodiment 1:

[0056] An elevator fault prediction method based on an improved slime mold algorithm to optimize the Elman neural network uses sensors to collect elevator operation data, and uploads the data to the background for data processing through a WIFI module. The background data processing uses an Elman neural network model optimized by an improved slime mold algorithm; the slime mold algorithm is optimized by sine-tent chaotic mapping and adding a new greedy step optimization, and outputs the optimization result with higher accuracy to the Elman neural network model, thereby optimizing the weights and thresholds of the Elman model, and finally outputting the prediction result; the specific operation steps include:

[0057] Step 1: Use sensors to collect various elevator operation data; elevator operation data includes: elevator shaft temperature, traction machine vibration, traction machine temperature, car vibration, elevator speed, elevator acceleration, elevator door opening and closing time, elevator door vibration, etc., represented by U(1)-U(8).

[0058] Step 2: Upload the collected data to the background for data preprocessing; the background for data preprocessing uses the Elman neural network model optimized by the improved slime mold algorithm; the slime mold algorithm is optimized by sine-tent chaotic mapping and adding a new greedy step. The added greedy step formula is as follows:

[0059]

[0060] Due to the lack of greedy steps, during the optimization process, the slime mold individual will move to that position regardless of whether the latest position is better than the current position, which will slow down the convergence speed of the algorithm during optimization. Therefore, adding a greedy step can optimize the optimization efficiency of the algorithm.

[0061] Formula (1) allows the slime mold individual to search near the current optimal value, compares the search result with the fitness of the current individual, and if the fitness value of the new individual is better than that of the current individual, updates the position to the new individual, otherwise retains the original position;

[0062] Among them, x n1 is a new individual, x best is the current optimal individual, y1 is the oscillation decay function, randA is a uniform random number in [-a, a], x r is a random individual in the entire population, x i is the current individual, t is the independent variable, and F is the fitness value.

[0063] The slime mold algorithm is optimized by sine-tent chaotic mapping, which increases the global search capability of the slime mold algorithm when updating the position. The formula for adding sine-tent chaotic mapping optimization to the slime mold algorithm is as follows:

[0064]

[0065] x n2 =x best +k n *(ω n x1-x2) (3)

[0066] In formula (2), k n is the sine-tent mapping in [0,1]; Formula (3) is the individual update formula of the slime mold algorithm after sine-tent chaos optimization; the individual update formula after adding chaos mapping has better global search performance and effectively jumps out of the local optimum; in formula (3), k n is a sine-tent mapping in [0,1], ω1 is the weight, x best is the current optimal individual, x1 and x2 are two random individuals.

[0067] Example:

[0068] Step 2.1: Select the test function Step 2.2: Determine the population size as N = 30, the number of iterations as T = 500, and the upper and lower bounds (lb, ub) = (-32, 32);

[0069] Step 2.3: Input the test function into the slime mold algorithm without the optimization of the sinusoidal chaos mapping method, and record the output individual values ​​of each generation;

[0070] Step 2.4: Input the test function into the slime mold algorithm optimized by sine chaos mapping, and record the individual values ​​of the output;

[0071] Step 2.5: Compare the two individual values ​​and draw conclusions. The results are shown in Table 1.

[0072]

[0073] Table 1

[0074] According to Table 1, the slime mold algorithm after sine-tent chaos optimization has better individual optimization ability. According to Table 1, the slime mold algorithm after sine-tent chaos optimization has better individual optimization ability than the original algorithm.

[0075] Step 3: For the data preprocessing in step 2, the collected data is normalized, the data is uniformly classified as [0,1], and input into the Elman neural network model.

[0076] The formula for the normalization process is as follows:

[0077]

[0078] In formula (4), x is the value of the current position, x' is the value after normalization, and x min is the minimum value in the data set, x max is the maximum value in the data set, c and d are the maximum and minimum values ​​after normalization, and the values ​​of c and d in formula (4) are [0,1].

[0079] Step 4: weights and thresholds of the initial Elman model;

[0080] Step 5: Use the slime mold algorithm to obtain the optimal value, use the sine-tent chaos optimized slime mold algorithm to search for individuals, and find the optimal individual as the optimal weight and threshold of the Elman neural network;

[0081] Through sorting, weight updating, and position updating, the slime mold algorithm can obtain an optimal individual, and input the optimal individual into the Elman model to optimize the weight or threshold of the model; the specific operation methods of the three parts of sorting, weight updating, and position updating are as follows:

[0082] S1. Sorting: sorting the fitness of the population to prepare for subsequent calculations.

[0083] S2. Weight update: When updating weights, different individuals have different update methods. The half of the population with better fitness uses formula (5), and the half with worse fitness uses formula (6).

[0084]

[0085] In formula (5) and formula (6), x best is the current optimal individual, x w It is the worst fitness value in this iteration. The half of the population with better fitness has a larger weight after the weight update formula. The better the fitness, the closer the weight of the individual is to 1, and vice versa.

[0086] S3. Position update: Update the position according to the formula. The update formula is:

[0087]

[0088] p = tanh(|F(x best )-F(x i )|) (9)

[0089] x n3 =rand*(x max -x min )+x min rand1≤z (10)

[0090]

[0091] x n5 =randB*x i rand1>z,rand2>p (12)

[0092] Formula (7) and formula (8) are parameter update formulas. As the number of iterations finally approaches 0, formula (9) is a threshold formula, which is used to decide whether to use formula (10) or formula (11) when updating the position. In formulas (7) and (8), a and b are values ​​that gradually iterate to 0, which are used to control the upper and lower limits of randA and randB, p is the judgment threshold, t is the current iteration number, and T is the maximum iteration number.

[0093] Formula (10)(11)(12) is a specific position update formula. Formula (10) is used to obtain a random position; Formula (11) searches near the current optimal position; and Formula (12) allows the individual to converge to 0.

[0094] In formulas (10), (11), and (12), x n3 、x n4 、x n5 is the latest position of the individual, rand1 and rand2 are random numbers in [0,1], z is a constant with a value of 0.03; randA is a uniform random number in [-a,a], and randB is a uniform random number in [-b,b].

[0095] Step 6: Determine whether the end condition is met. If the condition is met, assign the optimal value to the Elman neural network; if not, repeat step 5.

[0096] A portion of normalized elevator data UP(1)-UP(8), YP(1), and YP(2) are input into the Elman neural network as training data to train the Elman neural network. The specific training data is shown in Table 2:

[0097]

[0098] Table 2 Neural network training data

[0099] After the neural network training is completed, the test data U(1)-U(8) are input, and the output results are compared with the manually collected data Y(1) and Y(2) to determine the fault type. The prediction results are shown in Table 3.

[0100]

[0101] Table 3 Fault prediction results

[0102] In Table 3, U(1)-U(6) are the elevator shaft temperature, traction machine vibration, traction machine temperature, car vibration, elevator speed, and elevator acceleration, which are the main factors causing elevator car operation failures; U(7) and U(8) are the elevator door opening and closing time and elevator door vibration, which are the main factors causing elevator door failures. Y(1) is the elevator car operation failure; Y(2) is the elevator door operation failure. It can be seen from Table 2 that when U(1)-U(6) accounts for a large proportion, the output result is Y(1), that is, the elevator car operation failure; when U(7) and U(8) account for a large proportion, the output result is Y(2), that is, the elevator door failure. This shows that this method has a good prediction effect.

[0103] Step 7: According to the output results of the Elman neural network, determine whether the elevator will have a corresponding elevator fault; elevator faults can be divided into elevator car operation faults and elevator door faults, represented by Y(1) and Y(2).

[0104] The Elman neural network is a local feedback recursive neural network, which belongs to a nonlinear state space model. The topological structure of Elman can be divided into four layers: input layer, hidden layer, receiving layer and output layer. Figure 2 As shown in the figure, the role of the receiving layer is to receive the output of the hidden layer and pass the information corresponding to the previous time to the hidden layer in the form of feedback to establish a local ring structure; because the connection layer unit has a delayed memory effect on the contained characteristics, in the process of elevator fault prediction, the output value of the Elman neural network is more in line with the actual development trend of the data. The mathematical expression of the Elman neural network is:

[0105] C(j)=X(j-1) (13)

[0106] X(j)=f(ω1*U(j)+ω2*C(j)) (14)

[0107] Y(j)=g(ω3*X(j)) (15)

[0108] In formulas (13)(14)(15), C(j) is the feedback state vector; Y(j) is the output node vector; X(j) is the intermediate layer node unit vector; U(j) is the input vector; ω3 is the connection weight from the intermediate layer to the output layer; ω1 is the connection weight from the input layer to the intermediate layer; ω2 is the connection weight from the receiving layer to the intermediate layer; g() is the transfer function of the output neuron, which is a linear combination of the intermediate layer output; f() is the transfer function of the intermediate layer neuron, and the S function is often used.

Claims

1. An elevator fault prediction method based on an improved slime mold algorithm to optimize the Elman neural network, characterized in that: The elevator operation data is collected by sensors and uploaded to the background through the WIFI module for data processing. The background data processing adopts the Elman neural network model optimized by the improved slime mold algorithm. The slime mold algorithm is optimized by sine-tent chaotic mapping and adding new greedy steps to output the optimization results with higher accuracy to the Elman neural network model, thereby optimizing the weights and thresholds of the Elman model and finally outputting the prediction results. The specific operation steps include: Step 1: Use sensors to collect various elevator operation data; Step 2: Upload the collected data to the background for data preprocessing; the improved slime mold algorithm is to optimize the original slime mold algorithm by sine-tent chaotic mapping optimization and adding a new greedy step; The formula for the added greedy step is as follows: y1=e -t *sin(6t) t∈(0.5π) Formula (1) allows the slime mold individual to search near the current optimal value, compares the search result with the fitness of the current individual, and if the fitness value of the new individual is better than that of the current individual, updates the position to the new individual, otherwise retains the original position; Among them, x n1 is a new individual, x best is the current optimal individual, y1 is the oscillation decay function, randA is a uniform random number in [-a, a], x r is a random individual in the entire population, x i is the current individual, t is the independent variable, and F is the fitness value; The formula for adding sine-tent chaotic mapping optimization to the slime mold algorithm is as follows: x n2 =x best +k n *(ω n x1–x2) (3) In formula (2), k n is the sine-tent mapping in [0,1]; Formula (3) is the individual update formula of the slime mold algorithm after sine-tent chaos optimization; the individual update formula after adding chaos mapping has better global search performance and effectively jumps out of the local optimum; in formula (3), k n is a sine-tent mapping in [0,1], ω1 is the weight, x best is the current optimal individual, x1 and x2 are two random individuals; Step 3: For the data preprocessing in step 2, the collected data is normalized, the data is uniformly classified as [0,1], and input into the Elman neural network model; the normalization formula is as follows: In formula (4), x is the value of the current position, x' is the value after normalization, and x min is the minimum value in the data set, x max is the maximum value in the data set, c and d are the maximum and minimum values ​​after normalization, and the values ​​of c and d in formula (4) are [0,1]; Step 4: weights and thresholds of the initial Elman model; Step 5: Use the slime mold algorithm to obtain the optimal value; Through sorting, weight updating, and position updating, the slime mold algorithm obtains an optimal individual, and inputs the optimal individual into the Elman model to optimize the weight or threshold of the model; the specific operation methods of the three parts of sorting, weight updating, and position updating are as follows: S1. Sorting: sorting the fitness of the population to prepare for subsequent calculations; S2. Weight update: When updating weights, different individuals have different update methods. The half of the population with better fitness uses formula (5), and the half with worse fitness uses formula (6); In formula (5) and formula (6), x w is the worst fitness value in this iteration. The half of the population with better fitness has a larger weight after the weight update formula. The better the fitness, the closer the weight of the individual is to 1, and vice versa. S3. Position update: Update the position according to the formula. The update formula is: p=tanh(|F(x best )-F(x i )|) (9) x n3 =rand*(x max -x min )+x min rand1≤z (10) x n5 =randB*x i rand1>z,rand2>p (12) Formula (7) and Formula (8) are parameter update formulas. As the number of iterations finally approaches 0, Formula (9) is a threshold formula, which is used to decide whether to use Formula (10) or Formula (11) when updating the position. In Formulas (7) and (8), a and b are values ​​that gradually iterate to 0, which are used to control the upper and lower limits of randA and randB, p is the judgment threshold, t is the current iteration number, and T is the maximum iteration number. Formula (10)(11)(12) is a specific position update formula. Formula (10) is used to obtain a random position; Formula (11) searches near the current optimal position; Formula (12) allows the individual to converge to 0; In formulas (10), (11), and (12), x n3 、x n4 、x n5 is the latest position of the individual, rand1 and rand2 are random numbers in [0,1], z is a constant with a value of 0.03; randA is a uniform random number in [-a,a], and randB is a uniform random number in [-b,b]; Step 6: Determine whether the end condition is met. If the condition is met, assign the optimal value to the Elman neural network; if not, repeat step 5; Step 7: According to the output results of the Elman neural network, determine whether the elevator will have a corresponding elevator failure.

2. The elevator fault prediction method based on the improved slime mold algorithm and the Elman neural network optimization according to claim 1 is characterized in that: The elevator operation data described in step 1 include: elevator shaft temperature, traction machine vibration, traction machine temperature, car vibration, elevator speed, elevator acceleration, elevator door opening and closing time, elevator door vibration, represented by U(1)-U(8).

3. The elevator fault prediction method based on the improved slime mold algorithm and the Elman neural network optimization according to claim 1 is characterized in that: The elevator failure described in step 7 is divided into an elevator car operation failure and an elevator door failure, which are represented by Y(1) and Y(2).

4. The elevator fault prediction method based on the improved slime mold algorithm to optimize the Elman neural network according to claim 1 is characterized in that: The Elman neural network is a local feedback recursive neural network, which belongs to a nonlinear state space model. The topological structure of Elman is divided into four layers: input layer, hidden layer, receiving layer and output layer.

5. The elevator fault prediction method based on the improved slime mold algorithm and the Elman neural network optimization according to claim 4 is characterized in that: The function of the receiving layer is to receive the output of the hidden layer and pass the information corresponding to the previous time to the hidden layer in the form of feedback to establish a local ring structure; since the connection layer units have a delayed memory effect on the characteristics contained therein, in the process of elevator fault prediction, the output value of the Elman neural network is more in line with the actual development trend of the data.

6. The elevator fault prediction method based on the improved slime mold algorithm to optimize the Elman neural network according to claim 1 is characterized in that: The mathematical expression of the Elman neural network is: C(j)=X(j-1) (13) X(j)=f(ω1*U(j)+ω2*C(j)) (14) Y(j)=g(ω3*X(j)) (15) In formulas (13)(14)(15), C(j) is the feedback state vector; Y(j) is the output node vector; X(j) is the intermediate layer node unit vector; U(j) is the input vector; ω3 is the connection weight from the intermediate layer to the output layer; ω1 is the connection weight from the input layer to the intermediate layer; ω2 is the connection weight from the receiving layer to the intermediate layer; g() is the transfer function of the output neuron, which is a linear combination of the intermediate layer output; f() is the transfer function of the intermediate layer neuron, which uses the S function.

Images